TW536919B

TW536919B - Color image representing a patterned article

Info

Publication number: TW536919B
Application number: TW090117726A
Authority: TW
Inventors: Uri Gold; Eli Parente; Tally Gilat-Bernshtein; Edward Baranovsky; Tamir Margalit
Original assignee: Orbotech Ltd
Priority date: 1999-07-25
Filing date: 1999-08-03
Publication date: 2003-06-11
Also published as: US7200259B1; JP5086789B2; WO2001007893A2; JP2008164597A; US20070133862A1; TW512636B; IL131092A; JP2003511651A; IL131092A0; GB0201514D0; WO2001007893A3; JP4339541B2; GB2368120A; AU6012800A

Abstract

A method and device for determining a location of a border in a color image, the image comprising at least two color populations, between a first color region associated with a first one of the two color populations and a second color region associated with a second one of the two color populations, both the first color region and the second color region being comprised in the color image, comprising: identifying an approximate border location between said first color region and said second color region; determining a plurality of candidate border locations between the first color region and the second color region, each of said plurality of candidate border locations being determined by applying a corresponding border location method chosen from among a plurality of border location methods; choosing one method from among the plurality of border location methods as a preferred method; and determining a location of a border between said first color region and said second color region by designating one of said plurality of candidate border locations associated with the preferred method as the border.

Description

536919 A7 B7 五、發明説明（） 1 發明之領域本發明係關於供分析影像之裝置及方法，特別是，但不專指彩色影像，並尤指使用於檢查有圖案物體之影像分析裝置及方法。發明之背景供分析影像，特別是供使用於檢查有圖案物體之影像分析之裝置及方法，在此項技藝為熟知者。下列參考資料說明可使用於暸解本發明之影像處理方法： C. Gonzalez and P. Wintz，Digital Image Processing, Addison Wesley, Reading，MA，1987 ;以及536919 A7 B7 V. Description of the Invention (1) Field of the Invention The present invention relates to an apparatus and method for analyzing images, in particular, but not specifically to color images, and particularly to an image analysis device and method for inspecting patterned objects . BACKGROUND OF THE INVENTION Apparatuses and methods for analyzing images, especially for analyzing image objects for inspecting patterned objects, are well known in the art. The following references describe image processing methods that can be used to understand the present invention: C. Gonzalez and P. Wintz, Digital Image Processing, Addison Wesley, Reading, MA, 1987; and

John C. Russ，The Image Processing Handbook, CRC Press， 1994。下列參考資料說明可使用於暸解本發明之邊緣檢測方法： D. Marr and E. Hildreth, Theory of Edge Detection, Proceedings of the Royal Society of London ;以及 M. Chapron，"A new chromatic edge-detector used for color image segmentation”，11th APR International Conference on Pattern Recognition o 下列參考資料說明可使用於瞭解本發明之彩色影像分段方法： -4- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐） 536919 A7 B7 五、發明説明（John C. Russ, The Image Processing Handbook, CRC Press, 1994. The following reference materials illustrate the edge detection methods that can be used to understand the present invention: D. Marr and E. Hildreth, Theory of Edge Detection, Proceedings of the Royal Society of London; and M. Chapron, " A new chromatic edge-detector used for color image segmentation ", 11th APR International Conference on Pattern Recognition o The following reference materials explain the color image segmentation method that can be used to understand the present invention: -4- This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm) ) 536919 A7 B7 V. Description of the invention (

Philippe Pujas 及 Marie-Jose Aldon，’丨Robust Colour Image Segmentation’’，7th International Conference on Advanced Robotics， San Filiu de Guixols，Spain，1995 年 9 月 22 日，以及Philippe Pujas and Marie-Jose Aldon, ‘丨 Robust Colour Image Segmentation’ ’, 7th International Conference on Advanced Robotics, San Filiu de Guixols, Spain, September 22, 1995, and

Leila Shararenko，Maria Petrou，及 Josef Kittler, MAutomatic Watershed Segmentation of Randomly Textured Colour Images， IEEE。授予Lapidot之美國專利4,758,888號，說明一種供檢查沿一生產線行進之工作場所之方法及裝置，包括線上檢查瑕疵，而不中斷沿生產線之進行或工作場所。授予Katzir之美國專利5,058,982號，說明一種藉光學掃描器，特別可使用供檢查工作場所之照明系統。Leila Shararenko, Maria Petrou, and Josef Kittler, MAutomatic Watershed Segmentation of Randomly Textured Colour Images, IEEE. U.S. Patent No. 4,758,888 to Lapidot describes a method and apparatus for inspecting a workplace traveling along a production line, including inspecting defects online without interrupting progress or the workplace along the production line. U.S. Patent No. 5,058,982 to Katzir describes an optical scanner that is particularly useful for inspecting workplace lighting systems.

Orbotech Ltd.，Yavne，Israel 所擁有，1995 年 1 月 8 日所授予，並且名稱為"Illumination Device for an Optical Scanner” 之一以色列專利81450號，說明一種特別適合配合使用於自動檢查光學之掃描器使用，特別是供傳送強烈，多向光能至目標區之照明裝置。授予Aloni等人之美國專利5,586,058號，說明供檢查物體及檢測其中之缺陷之裝置及方法，包括檢查物體之二進制度級表示，檢查物體之灰度級表示，並較佳為重新檢查物體之灰階表示，以過濾錯誤警告及將缺陷分類。授予Caspi之美國專利5,774,572號，說明一種自動目視檢查系統，其較佳為可操作卷積物體之平面數位灰階影 -5- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐) ψ 裝訂Orbotech Ltd., owned by Yavne, Israel, granted on January 8, 1995, and named "Illumination Device for an Optical Scanner", an Israeli patent No. 81450, describing a scan that is particularly suitable for use with automatic inspection optics Device, especially for lighting devices that transmit strong, multi-directional light energy to the target area. US Patent No. 5,586,058 to Aloni et al. Describes devices and methods for inspecting objects and detecting defects therein, including inspection of objects System-level representation, gray-scale representation of inspection objects, and preferably gray-scale representation of re-inspection objects to filter error warnings and classify defects. US Patent No. 5,774,572 to Caspi describes an automatic visual inspection system that compares Plane digital grayscale image of a convolutional object that can be manipulated-5- This paper size applies to Chinese National Standard (CNS) A4 (210X297 mm) ψ binding

536919 A7 ____B7 五、發明説明（） 3 像’而以一相關於高斯函數（Gaussian function)之二次導數之濾光函數’形成具有記號值之平面卷積影像。求出在相鄰相反記號值間之零交叉，藉以達成邊緣在物體之地點。授予Caspi等人之美國專利5,774,573號，說明一種目視檢查系統’其使用物體之平面數位灰階影像之卷積，而以一相關於高斯函數（Gaussian function)之二次導數之濾光函數，形成具有記號值之平面卷積影像。Caspi等人之卷積，可利用一為正及一為負之高斯函數之差予以進行。 PCT申請案IL98/00393號，說明使用彩色檢查印刷電路板’包括使用彩色識別某些類型之狀況，諸如導體氧化作用。 PCT申請案IL98/00477號，說明供分析式表示影像，以便在非二進制像素可進行形態操作，諸如擴張，腐蝕，及比例測量之方法，較佳為較之使用先前方法更有效率。可使用於影像分析之方法，說明於下列刊物：536919 A7 ____B7 V. Explanation of the invention () 3 The image ′ is formed by a filter function related to the second derivative of the Gaussian function to form a plane convolution image with a mark value. Find the zero crossing between the adjacent opposite sign values to reach the edge where the object is. U.S. Patent No. 5,774,573 issued to Caspi et al. Describes a visual inspection system that uses the convolution of a planar digital grayscale image of an object and forms a filter function related to the second derivative of the Gaussian function to form Planar convolution image with marked value. The convolution of Caspi et al. Can be performed using the difference between a positive and a negative Gaussian function. PCT Application No. IL98 / 00393 states that the use of color to inspect printed circuit boards' includes the use of color to identify certain types of conditions, such as the oxidation of conductors. PCT Application No. IL98 / 00477 describes methods for representing images analytically so that morphological operations such as dilation, erosion, and scale measurement can be performed on non-binary pixels. It is preferable to be more efficient than using previous methods. Methods that can be used for image analysis are described in the following publications:

Dorin Comaniciu 及 Peter Meer， "Distribution Free Decomposition of Multivariate Data*1, SPR'98 Invited Submission, Department of Electrical and Computer Engineering, Rutgers University，Piscataway，NJ 08855, USA。以上所述及在整個本案說明書之所有參考資料，其揭示經予參考併入本案。發明之概述 -6- 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐) 536919 A7 B7 五、發明説明（） 4 本發明尋求提供一種改進之裝置及方法，供目視檢查物體，並分析其影像，特別是但不專指彩色影像。本發明之裝置及方法，據信為特別適用於分析一有圖案物體之影像，更特別適用於分析一有圖案物體之影像，其包括許多具有特徵光學屬性之元件。特別適合本發明之裝置及方法檢查之物體，包括球狀格柵陣列（ball grid array)基片 ("BGA”），印刷電路板基片，特別是包括多導體材料之印刷電路板基片，層壓印刷電路板，引線框架，平板顯示器，混合式晶片封裝基片，帶自動化接合基片，及其他類似多材料有圖案物體。所有先前技藝物體檢查系統，均有某些限制，使其僅可使用供檢查某些類型之物體，即使予以檢查之物體限於印刷電路板，積體電路，及類似電子組件之領域。本發明尋求提供分析影像，特別是彩色影像之改進裝置及方法，特別是供改進物體檢查系統之功能，增加系統所可檢查物體之差異性，並通常減低其限制之目的。改進彩色處理，據信在分析物體之多色影像可特別有效，特別是在多色物體之相鄰有色區域，具有彩色在單色影像難以彼此辨別之情形。例如，在檢查球狀格柵陣列基片或層壓印刷電路板之情形，可能宜於辨別在其表面為金之區域，及在其表面為銅之區域，以及辨別部份透明掩模所覆蓋之金屬區域及未覆蓋區域。同樣，在其表面為銀之本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐)Dorin Comaniciu and Peter Meer, " Distribution Free Decomposition of Multivariate Data * 1, SPR'98 Invited Submission, Department of Electrical and Computer Engineering, Rutgers University, Piscataway, NJ 08855, USA. All references mentioned above and throughout the description of this case, the disclosures of which are incorporated by reference into this case. Summary of the invention-6- This paper size applies the Chinese National Standard (CNS) A4 specification (210X 297 mm) 536919 A7 B7 V. Description of the invention () 4 The present invention seeks to provide an improved device and method for visual inspection of objects, And analyze its images, especially but not specifically color images. The apparatus and method of the present invention are believed to be particularly suitable for analyzing an image of a patterned object, and more particularly suitable for analyzing an image of a patterned object, and include many elements having characteristic optical properties. Objects particularly suitable for inspection by the apparatus and method of the present invention include a ball grid array substrate (" BGA "), a printed circuit board substrate, and particularly a printed circuit board substrate including a multi-conductor material , Laminated printed circuit boards, lead frames, flat panel displays, hybrid wafer package substrates, substrates with automated bonding, and other similar multi-material patterned objects. All prior art object inspection systems have certain limitations that make them Only certain types of objects can be used for inspection, even if the objects to be inspected are limited to the fields of printed circuit boards, integrated circuits, and similar electronic components. The present invention seeks to provide improved devices and methods for analyzing images, especially color images, especially It is for the purpose of improving the function of the object inspection system, increasing the variability of objects that can be inspected by the system, and generally reducing its limitations. Improved color processing is believed to be particularly effective in analyzing multicolor images of objects, especially in multicolor objects The adjacent colored areas have the situation that the color is difficult to distinguish each other in monochrome images. For example, in the inspection ball In the case of a grid-like grid array substrate or a laminated printed circuit board, it may be appropriate to discern areas where the surface is gold, areas where the surface is copper, and areas where metal is covered by a transparent mask. Covered area. Similarly, the size of the paper whose surface is silver applies the Chinese National Standard (CNS) A4 specification (210X297 mm)

裝訂 536919 A7 B7Binding 536919 A7 B7

區域可與其他區域予以辨別。使用本發明之方法及裝置，據信可較之在先前技藝更有效達成在此等區域間之辨別。Regions can be distinguished from other regions. Using the method and apparatus of the present invention, it is believed that discrimination between these areas can be achieved more effectively than in prior art.

裝訂在球狀格柵陣列基片之情形，僅意在作為實例，在可能宜於使用本發明之裝置及方法辨別或描述特徵之其他區域當中’另包括：裸基片，金屬塗層諸如在銅塾片及連接器上面之金鍍層；存在一至少部份透明塗層，諸如悍料掩模’覆蓋銅連接器，鍍金連接器或基片；不同物質，諸如銅連接器或鍍金連接器，位於一至少部份透明塗層，諸如烊料掩模下面；表面殘留物，諸如例如一種光敏抗蝕劑殘留物；在金屬塗層之空隙；表面不規律；擦痕；外來材料，汙點；以及銅連接器或金鍍層之氧化作用。也可能宜於辨別一區域諸如金屬連接器，金屬塗層或表面記號之形狀。請予察知，緊接上面所列舉之實例，係意在作為實例，而非限制，可能宜於辨別或特徵描述很多種類之區域。形成本發明之裝置及方法，據信具足夠變通性，使能在物體，諸如在檢查下之BGA，檢查及影像分析很多不同材料及材料組合。清予祭知，如以上所提及，在BGA之不同部位，通常將具有如以下所說明之不同彩色，彩色粒子數或其他光學特徵，並且因此辨別彩色及光學特徵，藉以可能辨別不同區域。在此一方面，請予特別察知，不同物質，包括不同金屬’及位於部份透明塗層，諸如焊料掩模下面之材料， -8 - 本紙張尺度適用家標A4規格(祕 536919 A7 五、發明説明（通常可使用本發明，根據其不同彩色及其以辨別及分類。特徵’予本㈣也尋求在處理單色影像，特別是在類似於以上關於彩色影像所述者之應用提供改進。、因此根據本發明之一種較佳實施例，提供-種方法， :在彩色影像，在一與二彩色粒子數之第一者關聯之第一心色4位與-與二彩色粒子數之第二者關聯之第二彩色部位之間，確定邊界之地點，影像包含至少二彩色粒子數，第一彩色邵位及第二彩色部位均包含在彩色影像中，包口 ·在第彩色邵位與第二彩色部位之間識別一近似邊界地點’在第一彩色部位與第二彩色部位之間確定許多候選邊界地點’許多候選邊界地點各係藉應用選自許多邊界地點方法當中之一對應邊界地點方法所確定；自許多邊界地點方法當中選擇一方法，作為一種較佳方法；以及指定與較佳方法關聯之許多候選邊界地點之一作為邊界，藉以在第一彩色邵位與第二彩色部位之間確定邊界之地點。另根據本發明之一種較佳實施例，彩色影像包括有圖案物品之影像。再根據本發明之一種較佳實施例，彩色影像包括電路之彩色影像。另外根據本發明之一種較佳實施例，彩色影像包括引線框架之彩色影像。 -9- 本紙張尺度適用中國國家標準(CNS) A4規格(21〇χ 297公釐) 裝訂 fBinding to a spherical grid array substrate is intended only as an example, among other areas where it may be appropriate to identify or describe features using the apparatus and methods of the present invention. 'In addition: bare substrates, metal coatings such as in Copper plating and gold plating on the connector; there is at least a part of a transparent coating, such as a mask to cover the copper connector, gold-plated connector or substrate; different substances, such as copper connector or gold-plated connector, Located under an at least partially transparent coating, such as a mask; surface residues such as, for example, a photoresist residue; voids in the metal coating; surface irregularities; scratches; foreign materials, stains; and Oxidation of copper connectors or gold plating. It may also be appropriate to identify the shape of an area such as a metal connector, metal coating, or surface marking. Please note that the examples listed immediately below are intended as examples, not limitations, and may be appropriate to identify or characterize many types of areas. The device and method of the present invention are believed to be sufficiently versatile to enable inspection and image analysis of many different materials and material combinations on objects, such as BGAs under inspection. Qingyu sacrifice, as mentioned above, in different parts of the BGA will usually have different colors, number of colored particles or other optical characteristics as explained below, and therefore distinguish between color and optical characteristics, thereby making it possible to identify different areas. In this regard, please be aware that different substances, including different metals, and materials that are partially transparent coatings, such as under the solder mask, -8-This paper size applies to the standard A4 (Secret 536919 A7 V. DESCRIPTION OF THE INVENTION (The present invention can generally be used to distinguish and classify according to its different colors and their characteristics. The feature 'Yu Benxi' also seeks to provide improvements in processing monochrome images, especially in applications similar to those described above with regard to color images. Therefore, according to a preferred embodiment of the present invention, a method is provided: in a color image, the first 4 colors of the first heart color associated with the first one of the number of two-color particles and the first number of the two-color particles The location of the boundary is determined between the two second color parts that are related to each other. The image contains at least two color particles. The first and second color parts are included in the color image. Identifying an approximate boundary location between the second color locations 'Determining many candidate boundary locations between the first color location and the second color location' Many candidate boundary locations are each borrowed by the application Determined from one of many boundary location methods corresponding to the boundary location method; selecting a method from among many boundary location methods as a better method; and designating one of the many candidate boundary locations associated with the better method as the boundary, thereby The location of the boundary between the first color shaw position and the second color part is determined. In addition, according to a preferred embodiment of the present invention, the color image includes an image of a patterned article. According to a preferred embodiment of the present invention, the color image A color image including a circuit. In addition, according to a preferred embodiment of the present invention, the color image includes a color image of a lead frame. -9- This paper size applies to China National Standard (CNS) A4 specification (21〇χ 297 mm) binding f

7 7 B7 五、發明說明（界^ ^根據本發明之—種較佳實施例，依據選擇較佳邊也-占万法’供預足彩色粒子數組合之規則，選擇較佳界地點方法。用|據本發明之一種較佳實施例，依據比較起因於應用至少二邊界地點方法之邊界地點，選擇較佳邊界地點方法。根據本發明之另„較佳實施例，也提供一種方法，供在先色〜像’在-與二彩色粒子數之第—者關聯之第一彩色4位與一與二彩色粒子數之第二者關聯之第二彩色部位 ' 崔足邊界之地點，影像包含至少二彩色粒子數，第彩色4位及第二彩色部位均包含在彩色影像中，該方法已括提供邊界分析圖，供每一對性質不同之彩色粒子數，匕括扣7JT —種較佳方法，供識別邊界在彩色粒子數間之地占識別在第一彩色部位與第二彩色部位間之近似邊界地 ”’占以及使用所指示之較佳方法，在邊界分析圖確定在第一彩色邵位與第二彩色部位間之邊界地點，供識別在第一彩色粒子數與第二彩色粒子數間之邊界地點。另根據本發明之一種較佳實施例，彩色影像包括有圖案物品之影像。再根據本發明之一種較佳實施例，彩色影像包括電路之彩色影像。另外根據本發明之一種較佳實施例，彩色影像包栝引 -10- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) 536919 A77 7 B7 V. Description of the invention (Boundary ^ ^ According to a preferred embodiment of the present invention, the method of selecting a better bounding point is selected according to the rule of selecting a better edge-occupying method for pre-sufficient color particle number combination. According to a preferred embodiment of the present invention, a better boundary location method is selected based on comparing boundary locations resulting from the application of at least two boundary location methods. According to another preferred embodiment of the invention, a method is also provided for Previous color ~ Image 'Place of the first color 4th place associated with the first of the two-colored particles and the second color portion associated with the second one of the two-colored particles' Cuizu boundary, the image contains The number of at least two colored particles, the fourth colored bit and the second colored part are included in the color image. The method has provided a boundary analysis chart for each pair of colored particles with different properties. A 7JT is better. Method for identifying the boundary between the number of colored particles and the approximate boundary between the first color part and the second color part, and using the indicated better method, determine the A boundary point between a color shaw bit and a second color part is used to identify a boundary point between the first color particle number and the second color particle number. In addition, according to a preferred embodiment of the present invention, the color image includes a patterned article According to a preferred embodiment of the present invention, the color image includes a color image of a circuit. In addition, according to a preferred embodiment of the present invention, the color image package is cited -10- This paper standard is applicable to the Chinese National Standard (CNS ) A4 size (210 X 297 mm) 536919 A7

線框架之彩色影像。電路包括印刷電而且根據本發明之一種較佳實施例路板。力根像尽發明之 " 〜丨玉·Γ入丨工只奶門，彩色影像包之以色影像’並且每—彩色粒子數與許多材料關聯。可种再根據本發明之一種較佳實施例’彩色影像也包括與第三彩色粒子數關聯之第三彩色部位，並且供彼此交々 …彩色部位’第二彩色部位’及第三彩色部位當中: 每二彩色部位’識別步驟包括在二彩色部位之間識別1 似邊界地點’及供每二彩色部位’確定步驟包括使用所葬示之較佳方法，在邊界分析圖在二彩色部位之間確定邊济之地點’供在與：彩色部位關聯之：彩色粒子數之間識别邊界地點。 t 另外根據本發明之一種較佳實施例，該方法也包括界足一有一窗口大小之窗口，檢查在窗口内之彩色影像之一部份，以確定性質不同之彩色部位之數，至少其一部份包括在該部份，並且如果性質不同之彩色部位之數大於二，將在窗口内之所有邊界分類為屬於一結點，並使用一種較佳方法，確定在窗口内之每一邊界之地點，供確定屬於一結點之邊界地點。而且根據本發明之一種較佳實施例，供確定屬於一結 -11 - 本紙張尺度適用中國國家標準(CNS) A4規4(210X297公着)_ 9 A7 點之邊界地點之軺另根據本發明'—匕括一種高水印方法。定特徵之彩色二㈣父佳實施例’使用有許多彩色界方法係選自示方法表示彩色影像，並且每-較佳每-彩色界定，並供在許多彩色界定特徵内之 -與彩色界定特徵關;/ :法包括-種方法，包括在法。彩色影像之組份進行邊緣定位方定月之—種較佳實施例’在-组份進行邊緣 'G括進行一種子像素輪廓元件定位方法。 ^卜根據本發明之_種較佳實施例，僅在—組份進行 k、足位方法，包括進行一種高水印方法。 :且根據本發明之—種較佳實施例，彩色影像表示方 / G括-# RGB方法，並且許多彩色界定特徵包括_ R 組份，-G组份，及一 B組份。另根據本發明之—種較佳實施例，彩色影像表示方法匕括種HSI方法’並且許多彩色界定特徵包括一 η組份’及S組份，以及一 I組份。根據本發明又另一較佳實施例，也提供一種方法，供自彩色影像產生一彩色形態圖，該方法包括提供彩色影像，縮小彩色影像以產生一包括許多彩色粒子數之縮小影像’在縮小影像確定在許多彩色粒子數之相鄰者間之邊界將知小影像分段以產生一包括邊界及至少一非邊界部 -12- 張尺度 A4i：4(210X297公⑻----—. 536919 A7 B7Color image of wire frame. The circuit includes a printed circuit board and a circuit board according to a preferred embodiment of the present invention. The invention of the force root image is fully invented. ~ 丨 Jade · Γ entry 丨 only milk door, color image package of color image ′ and the number of color particles per number is associated with many materials. According to a preferred embodiment of the present invention, the color image also includes a third color portion associated with the number of third color particles and intersects with each other ... among the color portion 'the second color portion' and the third color portion. : The step of identifying each two-colored part includes identifying 1 boundary-like locations between the two-colored parts, and the step of determining each two-colored part includes using a better method shown in the boundary analysis map between the two-colored parts. Determining the location of border areas is used to identify the boundary location between the number of colored particles and the number of colored particles. In addition, according to a preferred embodiment of the present invention, the method also includes defining a window having a window size, and inspecting a part of the color image in the window to determine the number of color parts with different properties, at least one of which The part is included in the part, and if the number of colored parts with different properties is greater than two, all the boundaries in the window are classified as belonging to a node, and a better method is used to determine the boundaries of each boundary in the window. Location for identifying a boundary location that belongs to a node. Furthermore, according to a preferred embodiment of the present invention, it is determined that it belongs to a knot -11-This paper size applies the Chinese National Standard (CNS) A4 Regulation 4 (210X297) _ 9 A7 of the boundary location of point A. According to the present invention '—Dagger includes a high watermark method. The color-defining embodiment of the color-defining feature of the second embodiment uses a number of color-boundary methods, which are selected from the method of representing color images, and each-preferably-color-defining, and for-color-defining features within many color-defining features. Off; /: Law includes-one method, including law. The component of the color image is determined by the edge positioning method. A preferred embodiment is a method of 'in-component edge', which includes a sub-pixel contour element positioning method. ^ According to a preferred embodiment of the present invention, the k-foot method is performed only on the component, including performing a high watermark method. : And according to a preferred embodiment of the present invention, the color image representation method / G includes-# RGB method, and many color defining characteristics include _ R component, -G component, and a B component. In addition, according to a preferred embodiment of the present invention, a color image representation method includes a HSI method 'and many color defining features include an n component' and an S component, and an I component. According to yet another preferred embodiment of the present invention, a method is also provided for generating a color morphology map from a color image. The method includes providing a color image and reducing the color image to generate a reduced image including a number of colored particles. The image determines the boundary between the neighbors of the number of colored particles. The small image is segmented to produce a boundary including at least one non-boundary portion. -12-scale A4i: 4 (210X297 cm. ---- 536919 A7 B7

五、發明説明（ 10 位 < 二進制影像，產生一至少一非邊界部位见★ / 木景）像，円采影像包括許多部份，以及指定一對應於彩色粒 a卞數之一之彩色識別至骨架之每一部份。另根據本發明之一種較佳實施例，指定步騾包括捽一彩色粒子數，為相對於彩色粒子數之骨架一部份之=疋地點之函數。工間再根據本發明之一種較佳實施例，指定步騾另勹骨架影像重疊在縮小影像。將另外根據本發明之一種較佳實施例，彩色影像包括一 24位元彩色影像。而且根據本發明之一種較佳實施例，縮小影像包括一 3 位元影像，並且許多彩色粒子數包括至多8彩色粒子數。另根據本發明之一種較佳實施例，確定步驟包括確定邊界至一像素之準確度。再根據本發明之一種較佳實施例，該方法也包括至少部份依據方法之結果，識別存在至少一形態特色。另外根據本發明之一種較佳實施例，彩色影像包括有圖案物品之影像。而且根據本發明之一種較佳實施例，有圖案物品包括電路。根據本發明之另一較佳實施例，也提供一種方法，供 -13-V. Description of the invention (10-bit < binary image, generating at least one non-boundary part, see ★ / wood scene) image, the mining image includes many parts, and a color recognition corresponding to one of the color particles a To every part of the skeleton. In addition, according to a preferred embodiment of the present invention, the specified step includes 捽 a number of colored particles, which is a function of 疋 location relative to a part of the skeleton relative to the number of colored particles. According to a preferred embodiment of the present invention, the designated step is another step. The skeleton image is superimposed on the reduced image. In addition, according to a preferred embodiment of the present invention, the color image includes a 24-bit color image. Furthermore, according to a preferred embodiment of the present invention, the reduced image includes a 3-bit image, and the number of colored particles includes a maximum of 8 colored particles. According to a preferred embodiment of the present invention, the determining step includes determining the accuracy of the boundary to a pixel. According to a preferred embodiment of the present invention, the method also includes at least partially identifying the existence of at least one morphological feature based on the result of the method. In addition, according to a preferred embodiment of the present invention, the color image includes an image of a patterned article. Moreover, according to a preferred embodiment of the present invention, the patterned article includes a circuit. According to another preferred embodiment of the present invention, a method is also provided for -13-

536919 A7536919 A7

☆ 〜像確足彩色輪廓元件（color element，簡稱cel )， =^去括提供彩色影像，縮小彩色影像以產生一包括許夕心色粒子數 < 縮小影像，確定在許多彩色粒子數之相鄰者1之邊界，邊界包括許多邊界分段，每一邊界分段被二心色粒子數所包81，並且供每-邊界分段，至少部份依據 G圍邊界刀奴之二彩色粒子數指定一值至邊界分段。另根據本發明之一種較佳實施例，彩色影像包一位元彩色影像。再根據本發明之一種較佳實施例，縮小影像包括一 3 位元影像。根據本發明之一種較佳實施例，確定步驟包括確足邊界至一像素之準確度。而且根據本發明之一種較佳實施例，確定步驟包足邊界至少於一像素之準確度。广另根據本發明之一種較佳實施例，每一邊界分段有一第-側及-第二側，並且指定步驟包括至少部份依：在邊界分段d侧之彩色粒子數及在邊界分段之第二侧色粒子數指定一值。 y 再根據本發明之一種較佳實施例，彩色影像包括案物品之影像。 ' 另外根據本發明之一種較佳實施例，有圖案物電路。栝☆ ~ like a color element (cel), = ^ to provide a color image, reduce the color image to produce a color image that includes the number of Xu Xixin particles < reduced image, determine the phase of many color particles The border of Neighbor 1, the border includes a number of border segments, each border segment is covered by the number of dichroic particles 81, and is provided for each-boundary segment, at least in part based on the number of colored particles Assign a value to the boundary segment. According to a preferred embodiment of the present invention, the color image includes a one-bit color image. According to a preferred embodiment of the present invention, the reduced image includes a 3-bit image. According to a preferred embodiment of the present invention, the determining step includes ensuring the accuracy of the boundary to one pixel. Moreover, according to a preferred embodiment of the present invention, the determining step includes the accuracy of the boundary being less than one pixel. According to a preferred embodiment of the present invention, each boundary segment has a first side and a second side, and the specifying step includes at least partly depending on: the number of colored particles on the d side of the boundary segment and the boundary analysis The number of colored particles on the second side of the segment specifies a value. y According to a preferred embodiment of the present invention, the color image includes an image of the item. 'In addition, according to a preferred embodiment of the present invention, there is a pattern object circuit. Gua

裝Hold

•14-• 14-

536919 A7536919 A7

根據本發明$ ,, ,, ^ 艾另一較佳實施例，也提供一種方法，供 1影像識別在平滑料之不規律，該方法包括線 :別”之許多點，有序之許多點各有至少一相鄭= 矣< P午夕點，在每一有序之許多點確定一至平滑曲線之法線方向、，並至少部份依據-局部法線差，包括在該點至平曲.泉之法線方向與在該點之至少一相鄰點至平滑曲線之法線方向間之差’識別有序之許多點之-點為不規律之一部份。另根據本發明之一種較佳實施例，影像包括有圖案物品之影像。再根據本發明之一種較佳實施例，影像包括電路之影像。、另外根據本發明〈一種較佳實施例，局部法線差包括在超過預定角度之方向之差。而且根據本發明之一種較佳實施例，預定角度包括一大約22度之角度。另根據本發明之一種較佳實施例，影像包括很多像素，並且有序之許多點各與一像素關聯。、，根據本發明之一種較佳實施例，在平滑曲線所包括之每一像素’係與有序之許多點之一關聯。另外根據本發明之一種較佳實施例，在平滑曲線所包括之僅有些像素，係與有序之許多點之一關聯。 -15- 本紙張尺度適财@ S家標準(CNS) Μ規格(21() χ 297公爱) 536919 A7 B7 五、發明説明（而且根據本發明之一種較佳實施例，選擇有些像素，以便不與有序之許多點之一關聯之固定數之像素，在每一有序 < 許多點與每一有序許多點之至少一近鄰者之間。另根據本發明之一種較佳實施例，該方法也包括確定供不規律之特色深度及特色孔徑，並至少部份依據確定步驟之結果將特色分類。再根據本發明之一種較佳實施例，分類包栝分類為一預定組類別之一，包括至少下列類別之一：一凹隙，一凸起部’及'輪廊。另外根據本發明之一種較佳實施例，該方法包括一種万法，供在電路檢測不規律，該方法包括在電路之影像，在一表7F二材料間之邊界之平滑曲線，根據該方法識別不規律。根據本發明之另一較佳實施例，也提供一種方法，供在物體 < 影像將不規律分類，該方法包括確定一供不規律 I特色深度及特色孔徑，並至少部份依據確定步驟之結果’將特色分類。另根據本發明之一種較佳實施例，物體包括電路。再根據本發明之一種較佳實施例，分類包栝分類為一預足組類別之一，包括至少下列類別之一：一凹隙，一凸起部，及一輪廓。根據本發明之另一較佳實施例，也提供一種方法，供 -16-According to another preferred embodiment of the present invention, a method is also provided for identifying the irregularity of the image in the smooth material. The method includes many points of line: other, and many points in order. Have at least one phase Zheng = 矣 < P midnight point, determine the normal direction of a smooth curve at each of many ordered points, and at least partially based on-local normal difference, including from this point to the flat curve The difference between the direction of the normal of the spring and the direction of the normal of at least one adjacent point to the smooth curve 'identifies a number of points that are ordered-the points are part of the irregularity. According to another aspect of the present invention, In a preferred embodiment, the image includes an image of a patterned article. According to a preferred embodiment of the present invention, the image includes an image of a circuit. In addition, according to the present invention (a preferred embodiment, the local normal difference is included in more than a predetermined The difference between the directions of the angles. According to a preferred embodiment of the present invention, the predetermined angle includes an angle of about 22 degrees. In addition, according to a preferred embodiment of the present invention, the image includes a plurality of pixels and an ordered number of points each With one pixel Correlation. According to a preferred embodiment of the present invention, each pixel included in the smooth curve is associated with one of the ordered points. In addition, according to a preferred embodiment of the present invention, the smooth curve Only some pixels are included, which are related to one of the many points in order. -15- The paper size is suitable for financial @ S 家标准 (CNS) Μ specifications (21 () χ 297 public love) 536919 A7 B7 V. Description of the invention (Also, according to a preferred embodiment of the present invention, some pixels are selected so that a fixed number of pixels that are not associated with one of the ordered many points are between each ordered < many points and each ordered many points At least one neighbor. According to a preferred embodiment of the present invention, the method also includes determining the characteristic depth and characteristic aperture for irregularity, and classifying the characteristics at least in part based on the result of the determining step. According to the present invention In a preferred embodiment, the classification is classified into one of a predetermined set of categories, including at least one of the following categories: a recess, a raised portion, and a wheel gallery. In addition, according to a preferred embodiment of the present invention, , The method includes a method for detecting irregularities in a circuit. The method includes smoothing a curve of a boundary between two materials in a table 7F and identifying irregularities according to the method. According to another preferred implementation of the present invention For example, a method is also provided for classifying irregularities in an object < image. The method includes determining a characteristic depth and characteristic aperture for the irregularity I, and classifying the characteristics based at least in part on the results of the determination step. According to a preferred embodiment of the present invention, the object includes a circuit. According to a preferred embodiment of the present invention, the classification is classified into one of a pre-foot group category, including at least one of the following categories: a recess, a protrusion According to another preferred embodiment of the present invention, a method is also provided for -16-

536919536919

在物體之影像之一部份識別缺陷，該部份靠近一在影像識別為邊緣之地點，該方法包括識別一與邊緣關聯之除外方向，並罪近識別為邊緣之地點，在許多不包括除外方向之方向搜尋缺陷。另根據本發明之一種較佳實施例，物體包括電路。再根據本發明之一種較佳實施例，除外方向包括一垂直於邊緣之方向。另外根據本發明之一種較佳實施例，缺陷包括擦痕。而且根據本發明之一種較佳實施例，缺陷包括汗點。根據本發明之另一較佳實施例，也提供一種方法，供在物體之影像識別表面缺陷，影像包括很多像素，該方法包括選擇一將行識別表面缺陷之部位，將該部位子取樣，藉以產生部位之子取樣影像，分析子取樣影像，並自其識別在該部位之表面缺陷。另根據本發明之一種較佳實施例，影像包括電路之彩色影像。再根據本發明之一種較佳實施例，子取樣包括將非相鄰像素子取樣。另外根據本發明之一種較佳實施例，相鄰像素各別之非相鄰像素之分隔，包括一子取樣比。而且根據本發明之一種較佳實施例，子取樣比在大約 113與大約ι49之間。 ___ - 17- 本紙張尺度適财^^(2igx挪公㈤裝訂Defects are identified in a portion of the object's image, which portion is near a location identified as an edge in the image. The method includes identifying an excluded direction associated with the edge and identifying the location of the edge as an edge, except in many exclusions. Direction to search for defects. According to a preferred embodiment of the present invention, the object includes a circuit. According to a preferred embodiment of the present invention, the excluded direction includes a direction perpendicular to the edge. In addition, according to a preferred embodiment of the present invention, the defects include scratches. Moreover, according to a preferred embodiment of the present invention, the defects include sweat spots. According to another preferred embodiment of the present invention, there is also provided a method for identifying a surface defect in an image of an object. The image includes many pixels. The method includes selecting a part that will identify the surface defect, and sampling the part, thereby Generate the sub-sampling image of the part, analyze the sub-sampling image, and identify the surface defect in the part from it. According to a preferred embodiment of the present invention, the image includes a color image of the circuit. According to a preferred embodiment of the present invention, subsampling includes subsampling non-adjacent pixels. In addition, according to a preferred embodiment of the present invention, the separation of adjacent pixels from non-adjacent pixels includes a sub-sampling ratio. Moreover, according to a preferred embodiment of the present invention, the sub-sampling ratio is between about 113 and about 49. ___-17- This paper is suitable for financial purposes ^^ (2igx Norway) Binding

536919536919

另根據本發日,, ，兔月 < 一種較佳實施例，子取樣包括根據名子取樣像素間之i斤々jp J_A 、又近似千均距離之子取樣。再根據本發明> # Α、月 < 一種較佳實施例，在子取樣像素間$ 近似平均距離，係在士系在大約5像素與大約11像素之間。另外根據本發日Ρ1 士 ^ 、、、 ^月又一種較佳實施例，子取樣步騾包招對該邵位之一部份子切卞取k ’並且使用該部位之許多部份或行子取樣步驟許多次。、而且根據本發明之_種較佳實施例，該部位之許多部份之至少二部份，包括該部位之部份重疊部份。根據本發明之另_查备、車义佳g她例，也提供一種方法，供應用一影像處理運算符至—在—像素陣列之像素，以產生 -二該方法包括提供一中心像素之nxn展開，如果中 ^ ^ 诼京私疋值0至一結果，如果中心像素不包括邊緣像素，如果一 1各東不疋邊緣像素，進扞下列 y驟，供η X n展開中之每一 .^ 4 母像素，如果一像素為邊緣像 =擇：像素作為選擇之像素，以及在…展開内之否目·]選：：像I疋邊緣像素’選擇鏡像作為選擇之像素，之乘f ^ ^ 象素，並將所選擇之像素值之宋知，及應用高斯運算符至中心 ^ 像常足結果加至和，並且和除以㈣之平均，以產生—結果。和並另根據本發明之一種較佳實施例，n等於3。根據本發明之另一較佳實 J也挺供一種方法，供According to the present day, a rabbit month < a preferred embodiment, the sub-sampling includes sub-sampling according to the name of the sub-sampling pixel jp J_A and the approximate mean distance. According to the present invention ># Α, Month < A preferred embodiment, the approximate average distance between sub-sampled pixels is about 5 pixels and about 11 pixels. In addition, according to another preferred embodiment of the present day P1, ^, ^, ^, the sub-sampling step includes cutting a part of the position to obtain k 'and using many parts or lines of the part. The subsampling step is performed many times. And, according to a preferred embodiment of the present invention, at least two parts of the many parts of the part include a part of the part overlapping the part. According to another example of the present invention, Che Yijia, also provides a method for supplying an image processing operator to pixels in an on-pixel array to generate-two. The method includes providing a central pixel of nxn Expand, if the middle ^ ^ 诼 Beijing private value is 0 to a result, if the center pixel does not include edge pixels, if a 1 does not include edge pixels, proceed to the following y steps for each of η X n expansion. ^ 4 female pixels, if a pixel is an edge image = select: pixel as the selected pixel, and whether it is in the expansion ...] select :: like I ： edge pixel 'select mirror as the selected pixel, multiply f ^ ^ Pixels, and the known value of the selected pixel value, and apply the Gaussian operator to the center ^ like the result of the full foot is added to the sum, and the sum is divided by the average of ㈣ to produce the result And in accordance with a preferred embodiment of the present invention, n is equal to three. According to another preferred embodiment of the present invention, J also provides a method for

裝訂Binding

-18 --18-

536919536919

以卷％《運算符計算梯度，包括提供影像處理運算符，成值之nxn p車列i形式’其中^為奇整數，及ηχη陣列、值可取在正值或負值，以具有正值之中心值應用景》像處理運算符，以Α , 以屋生弟一中間結果，以具有負值之中心值應用〜像處理運算符，以產生第二中間結果，並將第一中間結果及第二φ間έ士 4' < 一 Τ間〜果求和，以產生最後結果。斤另根據本發明之—種較佳實施例，該方法也包括比較第中間結果 < 算術符號及第二中間、结果之算術符號，並依據比較步驟之結果，確定存在局部極端值。再根據本發明之_種較佳實施例，確定包括以少於完全確實之確定。另外根據本發明之一種較佳實施例，η等於5。根據本發明之另一較佳實施例，也提供一種自動化光學秩查裝置，適合檢查有圖案物品，包括至少一檢測器，提供有冻少元件之有圖案物品之至少一部份之多色影像輪出，及處理電路，接收多色影像輸出，並提供至少下列檢查功能之一：在許多元件檢查一在至少一元件之金屬塗層，及在許多元件檢查至少一元件之至少部份透明塗層。根據本發明之另一較佳實施例，也提供一種自動化光莩叔查裝置，適合檢查有圖案物品，包括至少一檢測器，提供包括許多元件之有圖案物品之至少一部份之多色影像輸出，及處理電路，接收多色影像輸出，姐提供下列檢查 -19-Calculate the gradient with the operator "Volume%", including providing image processing operators, where the value is in the form of nxn p, where ^ is an odd integer, and the ηχη array. Application of central value image processing operator to A, to Yashengdi an intermediate result, to apply a central value with a negative value ~ to the image processing operator to produce a second intermediate result, and the first intermediate result and the first Two φ jian shi 4 '< one tsu kan ~ fruit sum to produce the final result. According to a preferred embodiment of the present invention, the method also includes comparing the first intermediate result < arithmetic symbol and the second intermediate and result arithmetic symbol, and determining the existence of a local extreme value based on the result of the comparison step. According to a preferred embodiment of the present invention, the determination includes a determination with less than complete certainty. In addition, according to a preferred embodiment of the present invention, n is equal to five. According to another preferred embodiment of the present invention, an automatic optical rank checking device is also provided, which is suitable for inspecting patterned articles, including at least one detector, and providing multi-color images of at least a part of patterned articles with few frozen components. The rotation and processing circuit receives the multi-color image output and provides at least one of the following inspection functions: inspecting a metal coating on at least one component on many components, and inspecting at least a part of transparent coating on at least one component on many components Floor. According to another preferred embodiment of the present invention, there is also provided an automated photo-inspection device suitable for inspecting a patterned article, including at least one detector, providing a multi-color image of at least a portion of a patterned article including a plurality of elements Output, and processing circuit, receiving multi-color image output, sister provides the following checks-19-

本紙張尺度適用中國國豕標準(CNS) Α4規格(210X 297公爱） 536919 A7 B7 五、發明説明（ 17 功能：在許多元件檢查一在至少一元件之金屬塗層，及在許多元件檢查在至少一元件之至少部份透明塗層。另根據本發明之一種較佳實施例，接收多色影像輸出 <處理電路提供下列另外之檢查功能：檢測在有圖案物品表面之殘留物。再根據本發明之一種較佳實施例，接收多色影像輸出之處理電路提供下列另外之檢查功能：檢測在有圖案物品表面之殘留物。另外根據本發明之一種較佳實施例，在許多元件之至少一元件檢查金屬塗層之功能，包括至少下列檢查模態之一 ·檢測在金屬塗層之空隙，檢測在金屬塗層之表面不規律’檢測在金屬塗層之擦痕，檢測在金屬塗層存在外來材料’檢測在金屬塗層之汙點，檢測金屬塗層之氧化作用，及檢測金屬塗層之形狀。而且根據本發明之一種較佳實施例，在許多元件之至少一元件檢查金屬塗層之功能，包括至少下列檢查模態之一 ·檢測在金屬塗層之空隙，檢測在金屬塗層之表面不規律’檢測在金屬塗層之擦痕，檢測在金屬塗層存在外來材料’檢測在金屬塗層之汙點，檢測金屬塗層之氧化作用，及檢測金屬塗層之形狀。另根據本發明之一種較佳實施例，在許多元件之至少一元件檢查至少部份透明塗層之功能，包括至少下列檢查 -20- 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐) 536919 A7This paper size applies to China National Standard (CNS) A4 specification (210X 297 public love) 536919 A7 B7 V. Description of the invention (17 Function: Inspection of many components-metal coating on at least one component, and inspection of many components on At least part of the transparent coating of at least one element. In addition, according to a preferred embodiment of the present invention, the receiving multi-color image output < processing circuit provides the following additional inspection functions: detecting residues on the surface of the patterned article. In a preferred embodiment of the present invention, the processing circuit for receiving multi-color image output provides the following additional inspection functions: detecting residues on the surface of a patterned article. In addition, according to a preferred embodiment of the present invention, at least many components A component inspects the function of the metal coating, including at least one of the following inspection modes: detection of voids in the metal coating, detection of irregularities in the surface of the metal coating 'detection of scratches in the metal coating, detection of metal coatings The presence of foreign materials' detects the stain on the metal coating, detects the oxidation of the metal coating, and detects the shape of the metal coating. According to a preferred embodiment of the present invention, the function of the metal coating is inspected on at least one of many components, including at least one of the following inspection modes. Regularly 'detect the scratches on the metal coating, detect the presence of foreign materials in the metal coating', detect the stains on the metal coating, detect the oxidation of the metal coating, and detect the shape of the metal coating. In a preferred embodiment, the function of at least a part of the transparent coating is checked on at least one of many components, including at least the following checks. -20- This paper size applies to China National Standard (CNS) A4 (210X 297 mm) 536919 A7

少部份透明塗層之汗點狀。模態之-：檢測在至少部份透明塗層之空隙，檢測在至少部份透明塗層之表面不規律，檢測在至少部份透明塗層之擦痕，檢測在至少部份透明塗層存在外來材料，檢測在至及檢測至少部份透明塗層之形再根據本發明之一種較佳實施例，在許多元件之至少 if件檢查至少部份透明塗層之功能，包括至少下列檢查模〜之檢測在至少部份透明塗層之空隙，檢測在至少部份透明塗層之表面不規律，檢測在i少部份透明塗層之擦痕，檢測在至少部份透明塗層存在外來材料，檢測在至 V邛伤透明塗層之汗點，及檢測至少部份透明塗層之形狀0 另外根據本發明之一種較佳實施例，在有圖案物品之表面檢測殘留物之功能，包括至少下列檢測模態之一：檢測在有圖案物品表面之表面不規律，檢測在有圖案物品表面存在外來材料，檢測在有圖案物品表面之汙點，及檢測在有圖案物品表面之記號之形狀。而且根據本發明之一種較佳實施例，在有圖案物品表面檢測殘留物之功能，包括至少下列檢測模態之一··檢測在有圖案物品表面之表面不規律，檢測在有圖案物品表面存在外來材料，檢測在有圖案物品表面之汙點，及檢測在有圖案物品表面之記號之形狀。 -21 - 本紙張尺度適用中國國家標準(CNS) A4規格(21〇x 297公釐)A small amount of sweaty spots on the transparent coating. Modal-: Detection of gaps in at least part of the transparent coating, detection of irregularities in the surface of at least part of the transparent coating, detection of scratches in at least part of the transparent coating, detection of presence of at least part of the transparent coating For foreign materials, test the shape of at least part of the transparent coating. According to a preferred embodiment of the present invention, check the function of at least part of the transparent coating on at least if of many components, including at least the following inspection molds ~ The detection is in the gaps of at least part of the transparent coating, the irregularity of the surface of at least part of the transparent coating, the scratches of a small part of the transparent coating, the presence of foreign materials in at least part of the transparent coating, Detects the perspiration point of the transparent coating that is scratched to V, and detects the shape of at least part of the transparent coating One of the detection modes: detection of irregularities on the surface of patterned articles, detection of foreign materials on the surface of patterned articles, detection of stains on the surface of patterned articles, and inspection There are marks in the shape of the surface of the article pattern. Furthermore, according to a preferred embodiment of the present invention, the function of detecting residues on the surface of a patterned article includes at least one of the following detection modalities: detecting irregularities on the surface of the patterned article, and detecting the presence of the surface of the patterned article For foreign materials, detect the stain on the surface of the patterned article, and detect the shape of the mark on the surface of the patterned article. -21-This paper size applies to China National Standard (CNS) A4 (21 × 297 mm)

裝訂 536919 A7 B7 五、發明説明（4 ) 19 另根據本發明之一種較佳實施例，有圖案物品包括電路。· 再根據本發明之一種較佳實施例，元件包括電路上之導體。另外根據本發明之一種較佳實施例，電路包括印刷電路板。而且根據本發明之一種較佳實施例，電路包括球狀格柵陣列基片。另根據本發明之一種較佳實施例，有圖案物品包含引線框架。再根據本發明之一種較佳實施例，金屬塗層包拾金屬鍍層。另外根據本發明之一種較佳實施例，元件包括球狀格柵陣列基片上之球體。而且根據本發明之一種較佳實施例，有圖案物品包括蚀刻金屬基片。另根據本發明之一種較佳實施例，有圖案物品包括壓花金屬基片。根據本發明之另一較佳實施例，也提供一種自動化光學檢查裝置，適合檢查物品，並包括至少一檢測器，提供物品之至少一部份之多色影像輸出，及處理電路，接收多色影像輸出，並提供邊緣檢測至子像素準確度，可操作在 -22- 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐) 536919 A7 B7Binding 536919 A7 B7 V. Description of the Invention (4) 19 According to a preferred embodiment of the present invention, the patterned article includes a circuit. According to a preferred embodiment of the present invention, the component includes a conductor on a circuit. In addition, according to a preferred embodiment of the present invention, the circuit includes a printed circuit board. Furthermore, according to a preferred embodiment of the present invention, the circuit includes a spherical grid array substrate. According to another preferred embodiment of the present invention, the patterned article includes a lead frame. According to a preferred embodiment of the present invention, the metal coating layer surrounds the metal plating layer. In addition, according to a preferred embodiment of the present invention, the element includes a sphere on a spherical grid array substrate. Furthermore, according to a preferred embodiment of the present invention, the patterned article includes an etched metal substrate. According to another preferred embodiment of the present invention, the patterned article includes an embossed metal substrate. According to another preferred embodiment of the present invention, an automated optical inspection device is also provided, which is suitable for inspecting an article and includes at least one detector, which provides multi-color image output of at least a part of the article, and a processing circuit that receives the multi-color Image output with edge detection to sub-pixel accuracy, can be operated at -22- This paper size applies to China National Standard (CNS) A4 specifications (210X 297 mm) 536919 A7 B7

五、發明説明（ 20 不同彩色之部位當中辨別邊界。另根據本發明之-種較佳實施例，處理電路為可操作在第-彩色部位與第二彩色部位之間識別近似邊界地點，並使用-種供在第-彩色粒子數與第二彩色粒子數之間描別邊界地點之較佳方法，確定在第—彩色部位與第二彩色部位間之邊界地點，藉以在―與—第—彩色粒子數關聯之V. Description of the invention (20 boundaries of different colored parts are distinguished. In addition, according to a preferred embodiment of the present invention, the processing circuit is operable to identify an approximate boundary point between the first-color part and the second-color part, and use -A better method for describing the boundary location between the first-color particle number and the second-color particle number, determining the boundary location between the first-color part and the second-color part, so that the -and-the-th color Particle number correlation

第-彩色部位與一與第二彩色部位關聯之第二彩色間之辨別邊界。再根據本發明之一種較佳實施例，多色影像輸出包括使用有許多彩色界定特徵之彩色影像表示方法所表示之彩色影像，並且每一較佳方法係選自許多現有方法，及許多見有方法包括一種方法’包括在與彩色界定特徵關聯之彩色〜像之組“進行邊緣定位方法，供在許多彩色界定特徵内之每一彩色界定特徵。另外根據本發明之一種較佳實施例，邊緣定位方法包括一種子像素輪廓元件定位方法。根據本發明之另一較佳實施例，也提供一種自動化光學檢查裝置，適合檢查物品，並包括至少_檢測器，提供 :有至少三不同部位之物品之至少一部份之影像輸出，該二部位各至少由一可藉檢測器檢測之光學特徵予以辨別，及處理電路，接收影像輸出，並提供邊緣檢測至子像素準確度，可操作辨別至少三不同部位之至少二者之邊界。卞The discrimination boundary between the first-color part and a second color associated with the second color part. According to a preferred embodiment of the present invention, the multi-color image output includes a color image represented by a color image representation method using a plurality of color-defining features, and each preferred method is selected from many existing methods and many common methods. The method includes a method 'including performing an edge localization method on a group of colors ~ images associated with the color-defining features for each color-defining feature within a plurality of color-defining features. In addition, according to a preferred embodiment of the present invention, the The positioning method includes a sub-pixel contour element positioning method. According to another preferred embodiment of the present invention, an automated optical inspection device is also provided, which is suitable for inspecting articles, and includes at least a detector, which provides: articles having at least three different parts At least a part of the image output, each of the two parts is identified by at least one optical feature that can be detected by the detector, and the processing circuit receives the image output and provides edge detection to sub-pixel accuracy, which can be identified by at least three operations. Boundary of at least two of different parts. 卞

A7A7

另根據本發明之一種較佳實施例，、，像輸出，並提供邊緣檢測：’接收影少三不同部位之邊界。子像素丰確度，可操作辨別至、再根據本發明之一種較佳實施例，處理電依據影像輸出提供測緣，並部位之材料。、相在母—有共同邊界之不同根據本發明之另-較佳實施例，也提供-種自動化光學檢查裝置，適合檢查物品，已栝至）一檢測器，提供部份之影像輸出，及處理電路，接收影像輸。至邵份依據影像輸出提供轉，其在有共同邊界 I 口口《至少一邵份之每一不同部位識別材料。另根據本發明之一種較伟余子像素準確度。 Mu例’處理電路可操作至二：：：明之另一較佳實施例，也提供-種自動化光子“裝置，週合檢查物品，包括至少—檢測器，提供物品（至少-邵份之多色影像輸出，處理電路，接收多色影像輸出，並提供至少下列檢查功能之—：乡色影像之分段為一包括許多彩色部位，各有許多代表性彩色可能性之一《分段彩色影像，分段彩色影像之二進制化為一包括數部位及邊界之二進制影像’該等部位之形態腐触以提供一表 -諸部位之骨架’以及指定彩色部位之至骨架中之諸元件。 536919 A7 B7 五、發明説明（） 22 另根據本發明之一種較佳實施例，處理電路可操作使骨架及分段彩色影像重疊，及指定表示該等部位之資料至骨架中之諸元件為重疊之函數。再根據本發明之一種較佳實施例，處理電路可操作，在一彩色部位，在骨架之一元件之重疊地點，指定表示該等部位之資料至骨架中之諸元件。根據本發明之另一較佳實施例，也提供一種自動化光學檢查裝置，適合檢查物品，包括至少一檢測器，提供物品之至少一部份之多色影像輸出，及處理電路，接收多色影像輸出，及可操作將多色影像分段為選自預先界定數之彩色部位可能性當中之許多彩色部位，並產生多色影像之表示，包括彩色部位資料，表示影像中之每一彩色部位，及邊界資料供表示彩色部位當中之每一邊界。根據本發明之再一較佳實施例，也提供表示有圖案物品之彩色影像，彩色影像包括形態腐蝕有圖案物品之多色影像。另根據本發明之一種較佳實施例，形態腐蝕多色影像之部份之彩色，指示在有圖案物品之各不同部位之材料。再根據本發明之一種較佳實施例，有圖案物品包括電路。另外根據本發明之一種較佳實施例，元件包括電路上之導體。 -25- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) 23 五、發明説明（而且根據本發明士 _ 較佳實施例，電路包括印刷電路板。另根據本發明夕—括^ ^ 〈種較佳實施例，電路包括球狀格In addition, according to a preferred embodiment of the present invention, the image output is provided, and edge detection is provided: 'receiving the boundaries of three different parts of the image. The sub-pixel abundance is operable to discriminate to, and then according to a preferred embodiment of the present invention, the processing material is used to provide the edge and the part according to the image output. Phases in the mother—the differences that have a common boundary. According to another preferred embodiment of the present invention, an automated optical inspection device is also provided, suitable for inspecting items, and has reached to) a detector that provides part of the image output, and The processing circuit receives image input. To Shao Fen provides the turn based on the image output, which identifies the material at each different part of at least one Shao Fen with a common boundary I. In addition, according to the present invention, a more accurate sub-pixel accuracy is obtained. The Mu 'processing circuit can be operated to two ::: another preferred embodiment of the Ming, also provides-an automated photon "device, to check the items, including at least-detectors, to provide items (at least-Shaofen multicolor Image output, processing circuit, receiving multi-color image output, and providing at least the following inspection functions: the segmentation of the country-color image is one including many color parts, each with many representative color possibilities The binarization of the segmented color image is a binary image that includes several parts and boundaries. The shapes of these parts are corroded to provide a table-the skeleton of the parts, and the components of the designated color parts to the skeleton. 536919 A7 B7 V. Description of the invention (22) According to a preferred embodiment of the present invention, the processing circuit is operable to overlap the skeleton and the segmented color image, and designate the components representing the data of these parts into the skeleton as a function of the overlap. According to a preferred embodiment of the present invention, the processing circuit is operable to designate a table in a color portion at an overlapping position of one element of the skeleton. According to another preferred embodiment of the present invention, an automatic optical inspection device is also provided, which is suitable for inspecting articles, including at least one detector, and providing at least a part of the articles. Multi-color image output, and processing circuit, receiving multi-color image output, and operable to segment the multi-color image into many color parts selected from the possibility of color parts of a predefined number, and generate a multi-color image representation, including The color part data represents each color part in the image, and the boundary data is used to represent each boundary in the color part. According to still another preferred embodiment of the present invention, a color image representing a patterned article is also provided. The color image includes Morphologically eroded multi-color images of patterned articles. In addition, according to a preferred embodiment of the present invention, the color of the morphologically corroded multi-colored images indicates materials at different parts of the patterned article. In a preferred embodiment, a patterned article includes a circuit. In addition, according to a preferred embodiment of the present invention, a component package Including the conductors on the circuit. -25- This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) 23 V. Description of the invention (and according to the invention_ preferred embodiment, the circuit includes a printed circuit board In addition, according to the present invention—including ^ ^ <a preferred embodiment, the circuit includes a spherical grid

陣列基片。 W 再根據本發明> _ ^ 又一種較佳貫施例，有圖案物品包括引線框架。Array substrate. W According to the present invention > _ ^ yet another preferred embodiment, the patterned article includes a lead frame.

另外根據本發明> _ iA 月义種較佳實施例，金屬塗層包括金屬鍍層。而且根據本發明> _ 種較佳貫施例，元件包括球狀格柵陣列基片上之球體。另根據本發明〈一種較佳實施例，有圖案物品刻金屬基片。再根據本發明之一種較佳余者R安仏 4平乂住K她例，有圖案物品包括壓花金屬基片。根據本發明之另一較佳實施例，也提供一種自動化光 :檢查裝置，適合檢查物品，包括至少一檢測器，提供物 P <至/部份 < 多色影像輸出，及處理電路，接收多色 :像鈿出’並提供表示物品之彩色影像，¥色影像包括物品之至少一部份之形態腐蝕多色影像。、另根據本發明之-種較佳實施例，形態腐！i多色影像之省4饧之彩色，指π在物品之各不同部位之材料。根據本發明〈另一較佳實施例，也提供有圖案物品之 -26- 536919 A7 B7 五、發明説明（） 24 彩色影像表示，彩色影像包括一圖，其指示在物品之不同部位間之邊界，並在有共同邊界之每一不同部位識別材料。另根據本發明之一種較佳實施例，圖之諸部份之彩色指示在不同部位之材料。再根據本發明之一種較佳實施例，每一彩色包括一自預定彩色可能性之彩色。另外根據本發明之一種較佳實施例，有圖案物品包括電路。而且根據本發明之一種較佳實施例，元件包括電路上之導體。另根據本發明之一種較佳實施例，電路包括印刷電路板。再根據本發明之一種較佳實施例，電路包括球狀格柵陣列基片。另外根據本發明之一種較佳實施例，有圖案物品包括引線框架。而且根據本發明之一種較佳實施例，金屬塗層包括金屬鍍層。另根據本發明之一種較佳實施例，元件包括球狀格栅陣列基片上之球體。再根據本發明之一種較佳實施例，有圖案物品包括蝕 -27- 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐） W6919 A7In addition, according to a preferred embodiment of the present invention > iA, the metal coating layer includes a metal plating layer. And according to a preferred embodiment of the present invention, the element includes a sphere on a spherical grid array substrate. According to another preferred embodiment of the present invention, a patterned article is engraved with a metal substrate. According to another preferred embodiment of the present invention, R 仏仏乂乂乂她她她例 example, the patterned article includes an embossed metal substrate. According to another preferred embodiment of the present invention, there is also provided an automated light: inspection device suitable for inspecting articles, including at least one detector, providing objects P < to / partially, < multi-color image output, and processing circuit, Receiving multi-color: like 钿出 'and provide a color image representing the item. The ¥ -color image includes at least a part of the item in the form of corroded multi-color image. In addition, according to a preferred embodiment of the present invention, the form is rotten! The color of the multi-color image of the province of 4 饧 refers to the material of π in different parts of the article. According to the present invention <another preferred embodiment, -26- 536919 A7 B7 is also provided with a patterned article V. Description of the invention () 24 Color image representation, the color image includes a picture indicating the boundary between different parts of the article And identify the material at each different location with a common boundary. In addition, according to a preferred embodiment of the present invention, the colors of the parts of the drawing indicate the materials in different parts. According to a preferred embodiment of the present invention, each color includes a color from a predetermined color possibility. In addition, according to a preferred embodiment of the present invention, the patterned article includes a circuit. Moreover, according to a preferred embodiment of the present invention, the component includes a conductor on a circuit. According to a preferred embodiment of the present invention, the circuit includes a printed circuit board. According to a preferred embodiment of the present invention, the circuit includes a spherical grid array substrate. In addition, according to a preferred embodiment of the present invention, the patterned article includes a lead frame. Furthermore, according to a preferred embodiment of the present invention, the metal coating layer includes a metal plating layer. According to another preferred embodiment of the present invention, the element comprises a sphere on a spherical grid array substrate. According to a preferred embodiment of the present invention, the patterned article includes an etch. -27- This paper size applies the Chinese National Standard (CNS) A4 specification (210X 297 mm) W6919 A7

刻金屬基片。另外根據本發明之一種較佳實施例，有圖案物品包括壓花金屬基片。根據本發明之另一較佳實施例，也提供一種自動化光學檢查裝置，適合檢查物品，包括至少一檢測器，提供物 <至少一部份之多色影像輸出，及處理電路，接收多色影像輸出，並提供表示物品之彩色影像，彩色影像包括一圖’其指不在物品之不同部位間之邊界，並在有共同邊界之每一不同部位識別材料。另根據本發明之一種較佳實施例，圖之諸部份之彩色指示在物品之各不同部位之材料。再根據本發明之一種較佳實施例，處理電路，接收影像輸出’並提供邊界指示至子像素準確度。另外根據本發明之一種較佳實施例，每一彩色包括選自預定彩色可能性當中之彩色。而且根據本發明之一種較佳實施例，有圖案物品包括電路。另根據本發明之一種較佳實施例，電路包括印刷電路板。再根據本發明之一種較佳實施例，電路包括球狀格柵陣列基片。另外根據本發明之一種較佳實施例，有圖案物品包括 -28-Carved metal substrate. In addition, according to a preferred embodiment of the present invention, the patterned article includes an embossed metal substrate. According to another preferred embodiment of the present invention, an automatic optical inspection device is also provided, which is suitable for inspecting articles, including at least one detector, providing at least a part of a multi-color image output, and a processing circuit for receiving multi-color The image is output and provides a color image representing the item. The color image includes a picture which indicates that the material is not at the boundary between different parts of the article and is identified at each different part with a common boundary. In addition, according to a preferred embodiment of the present invention, the colors of the parts of the drawing indicate the materials at different parts of the article. According to a preferred embodiment of the present invention, the processing circuit receives the image output 'and provides a boundary indication to the sub-pixel accuracy. In addition, according to a preferred embodiment of the present invention, each color includes a color selected from among predetermined color possibilities. Moreover, according to a preferred embodiment of the present invention, the patterned article includes a circuit. According to a preferred embodiment of the present invention, the circuit includes a printed circuit board. According to a preferred embodiment of the present invention, the circuit includes a spherical grid array substrate. In addition, according to a preferred embodiment of the present invention, the patterned article includes -28-

本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐） 536919 A7 B7 五、發明説明（引線框架。而且根據本發明之一種較佳實施例，不同材料包括不同金屬。另根據本發明之一種較佳實施例，不同部位包括未塗布金屬導體，塗布金屬導體，未塗布基片，及塗布基片。再根據本發明之一種較佳實施例，有圖案物品包括蝕刻金屬基片。另外根據本發明之一種較佳實施例，有圖案物品包括壓花金屬基片。根據本發明之另一較佳實施例，也提供一種自動化光學檢查裝置，適合檢查有圖案物品，包括至少一檢測器，提供包括至少三材料有可檢測邊緣之有圖案物品之至少一部份之影像輸出，及處理電路，接收影像輸出，並提供檢查功能，包括影像輸出之二進制化，以提供一包括部位影像資料及邊界影像資料之二進制影像。另根據本發明之一種較佳實施例，有圖案物品包括電路。再根據本發明之一種較佳實施例，電路包括印刷電路板。另外根據本發明之一種較佳實施例，印刷電路板包括球狀格柵陣列基片。而且根據本發明之一種較佳實施例，有圖案物品包括 -29- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) 裝訂This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) 536919 A7 B7 V. Description of the invention (lead frame. And according to a preferred embodiment of the present invention, different materials include different metals. According to this According to a preferred embodiment of the present invention, different parts include an uncoated metal conductor, a coated metal conductor, an uncoated substrate, and a coated substrate. According to a preferred embodiment of the present invention, the patterned article includes an etched metal substrate. In addition, according to a preferred embodiment of the present invention, the patterned article includes an embossed metal substrate. According to another preferred embodiment of the present invention, an automated optical inspection device is also provided, which is suitable for inspecting the patterned article, including at least one inspection. The device provides image output including at least a part of a patterned article having at least three materials with detectable edges, and a processing circuit that receives the image output and provides inspection functions, including binarization of the image output, to provide an image including the part Binary image of data and boundary image data. Another preferred implementation of the present invention For example, a patterned article includes a circuit. According to a preferred embodiment of the present invention, the circuit includes a printed circuit board. In addition, according to a preferred embodiment of the present invention, the printed circuit board includes a spherical grid array substrate. In a preferred embodiment of the present invention, the patterned articles include -29- This paper size is applicable to China National Standard (CNS) A4 (210 X 297 mm) binding

536919 A7 B7 五、發明説明（引線框架。另根據本發明之一種較佳實施例，有圖案物品包括蝕刻金屬基片。再根據本發明之一種較佳實施例，有圖案物品包括壓花金屬基片。根據本發明之另一較佳實施例，也提併一種自動化光學檢查裝置，適合檢查有圖案物品，包括至少一檢測器，提供包括許多材料，在許多材料之間有可檢測邊界之有圖案物品之至少一部份之影像輸出，及處理電路，接收影像輸出’並提供檢查功能，包括關於邊界之圖案分析，以提供沿邊界存在凹隙及凸起部之指示。另根據本發明之一種較佳實施例，有圖案物品包括電路。再根據本發明之一種較佳實施例，有圖案物品包括印刷電路板。、另外根據本發明之一種較佳實施例，印刷電路板包括球狀格撕陣列基片。而且根據本發明之-種較佳實施例，有圖案物品包接引線框架。一另根據本發明之-種較佳實施例，許多材料包括至少三材料。再根據本發明之一種較佳實施例，材料包括自下 I__ -30- 本紙張尺歧财關家^^ 列類536919 A7 B7 V. Description of the invention (lead frame. In addition, according to a preferred embodiment of the present invention, the patterned article includes an etched metal substrate. According to a preferred embodiment of the present invention, the patterned article includes an embossed metal substrate According to another preferred embodiment of the present invention, an automated optical inspection device is also incorporated, which is suitable for inspecting a patterned article, including at least one detector, and providing a plurality of materials including a detectable boundary between the plurality of materials. The image output of at least a part of the patterned article, and the processing circuit, receives the image output 'and provides inspection functions, including pattern analysis on the boundary to provide an indication of the presence of pits and protrusions along the boundary. In a preferred embodiment, the patterned article includes a circuit. According to a preferred embodiment of the present invention, the patterned article includes a printed circuit board. In addition, according to a preferred embodiment of the present invention, the printed circuit board includes a ball grid. The array substrate is torn. According to a preferred embodiment of the present invention, a patterned article is wrapped around the lead frame. According to a preferred embodiment of the present invention, many materials include at least three materials. According to a preferred embodiment of the present invention, the materials include the following materials: I__ -30

裝 536919Equipment 536919

組之材料：未塗布金屬導體片，塗布基片。塗布材料導體，未塗布基另外根據本發日月> ,,, 月 < 一種較佳實施例，有圖案物品包括蝕刻金屬基片。而且本發明金屬基片。之一種較佳實施例，有圖案物品包括壓花Group materials: uncoated metal conductor sheet, coated substrate. Coated material conductor, uncoated base In addition, according to the present day, month, and month, a preferred embodiment, the patterned article includes an etched metal substrate. And the present invention is a metal substrate. A preferred embodiment, the patterned article includes embossing

裝、：根據本發明之一種較佳實施例，其中包括圖案分析查功能’包括沿關於邊界之影像資料所界定之邊界，識別有：之許多點，有序之許多點各有至少一相鄰點在有、、” 在母有序之許多點確定至平滑曲線之法線万向，並至少部份依據一局部法線差，包括在該點至平滑曲、泉〈法線方向與在該點之至少一相鄰點至平滑曲線之法訂線万向間之差’識別有序之許多點之-點為不規律之-部份。 t 庳=據本發明之另一較佳實施例，也提供一種自動化光予衩且农置，適合檢查物品，包括至少一檢測器，提供一包括多材料在材料之間有可檢測邊界之物品之至少一部份之以像知出，及處理電路，接收影像輸出，並提供至少下列禾、且功旎之一：影像輸出之分段為一包括部位影像資料及邊界影像資料之分段影像，使用一第一技術分析部位影像貝料，及使用一不同於第一技術之第二技術分析邊界影像資料。 31 - 536919 A7 B7 五發明説明（ 29 另根據本發明之一種較佳實施例，物品包括電路。再根據本發明之一種較佳實施例，第一技術包括近鄰刀析及第一技術包括邊界依循分析，供自非靠近邊界像素區別靠近邊界像素。根據本發明之另一較佳實施例，也提供一種自動化光子禾X查裝置’適合檢查物品，包括至少一檢測器，提供一包括多材料，在材料之間有可檢測邊界之物品之至少一部份之影像輸出，及處理電路，接收多色影像輸出，並提供至少下列檢查功能之一：多色影像輸出之分段為一包括選自一組預定彩色當中之彩色部位之分段彩色影像，使用一第一技術分析關於第一彩色之部位影像資料，及使用一不同於第一技術之第二技術分析關於第二彩色之部位影像資料。 / 、另根據本發明之一種較佳實施例，物品包括電路。再根據本發明之一種較佳實施例，電路包括印刷電路板。另外根據本發明之一種較佳實施例，印刷電路板包括球狀格柵陣列基片。而且根據本發明之一種較佳實施例，物品包括引線框架。在：據本發明之另-較佳實施例，也提供-種自動化光子一裝置，適合檢查物品，包括至少一檢測器，提供一 -32 536919 A7 B7 五、發明説明（） 30 包括多材料，在材料之間有可檢測邊界之物品之至少一部份之影像輸出，及處理電路，接收影像輸出，並提供至少下列檢查功能之一：影像輸出之分段為一包括部位影像資料及邊界影像資料之分段影像，並以一種不受至少有些邊界影像資料影響之方式，使部位影像資料平滑。另根據本發明之一種較佳實施例，物品包括電路。再根據本發明之一種較佳實施例，電路包括印刷電路板。另外根據本發明之一種較佳實施例，印刷電路板包括球狀格柵陣列基片。而且根據本發明之一種較佳實施例，物品包括引線框架。根據本發明之另一較佳實施例，也提供一種自動化光學檢查裝置，適合檢查電路，包括至少一檢測器，提供包括多材料之電路之至少一部份之影像輸出，及處理電路，接收影像輸出，並提供存在至少一預定材料之輸出指示。另根據本發明之一種較佳實施例，電路包括印刷電路板。再根據本發明之一種較佳實施例，印刷電路板包括球狀格柵陣列基片。另外根據本發明之一種較佳實施例，電路包括平板顯示器基片。 -33- 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐) 裝訂Equipment: According to a preferred embodiment of the present invention, it includes a pattern analysis and search function, which includes a boundary defined by the image data about the boundary, and identifies: a plurality of points, and an ordered number of points each having at least one adjacent The points in the ,, "" are determined at a number of points in the mother order to the normal of the smooth curve, and are based at least in part on a local normal difference, including the smooth curve, the normal direction and the The difference between at least one adjacent point of the point and the normal of the normal line of the smooth curve 'identifies a number of points that are ordered-points are irregular-part. T 庳 = according to another preferred embodiment of the present invention , Also provides an automatic light to the farm, suitable for inspecting items, including at least a detector, providing at least a part of an item including a plurality of materials with a detectable boundary between the materials, as well as image processing, and processing A circuit that receives image output and provides at least one of the following, and functions: the segmentation of the image output is a segmented image that includes part image data and boundary image data, using a first technique to analyze the part image shell material, and Make A second technique different from the first technique is used to analyze the boundary image data. 31-536919 A7 B7 Five invention descriptions (29 In addition, according to a preferred embodiment of the present invention, the article includes a circuit. Then according to a preferred implementation of the present invention For example, the first technique includes near-neighbor analysis and the first technique includes boundary-following analysis, which is used to distinguish near-boundary pixels from non-close-to-boundary pixels. According to another preferred embodiment of the present invention, an automated photon and X-check device is also provided. Suitable for inspecting articles, including at least one detector, providing an image output including multiple materials, at least a part of the article with a detectable boundary between the materials, and a processing circuit, receiving multi-color image outputs, and providing at least the following inspections One of the functions: The segmentation of the multi-color image output is a segmented color image including color parts selected from a set of predetermined colors, using a first technique to analyze the image data of the part in the first color, and using a different The second technique of the first technique analyzes the image data of the part of the second color. In the embodiment, the article includes a circuit. According to a preferred embodiment of the present invention, the circuit includes a printed circuit board. In addition, according to a preferred embodiment of the present invention, the printed circuit board includes a spherical grid array substrate. In a preferred embodiment of the invention, the article includes a lead frame. In accordance with another preferred embodiment of the invention, an automated photon device is also provided, suitable for inspecting articles, including at least one detector, providing -32 536919 A7 B7 V. Description of the invention () 30 Includes multiple materials, image output of at least a part of the article with a detectable boundary between the materials, and a processing circuit that receives the image output and provides at least one of the following inspection functions: image The output segment is a segment image that includes part image data and boundary image data, and smoothes the part image data in a way that is not affected by at least some of the boundary image data. According to another preferred embodiment of the present invention, the article includes a circuit. According to a preferred embodiment of the present invention, the circuit includes a printed circuit board. In addition, according to a preferred embodiment of the present invention, the printed circuit board includes a spherical grid array substrate. Moreover, according to a preferred embodiment of the present invention, the article includes a lead frame. According to another preferred embodiment of the present invention, there is also provided an automated optical inspection device suitable for inspecting a circuit, including at least one detector, providing an image output of at least a part of a circuit including multiple materials, and a processing circuit to receive an image Output and provide an output indication of the presence of at least one predetermined material. According to a preferred embodiment of the present invention, the circuit includes a printed circuit board. According to a preferred embodiment of the present invention, the printed circuit board includes a spherical grid array substrate. In addition, according to a preferred embodiment of the present invention, the circuit includes a flat panel display substrate. -33- This paper size applies to China National Standard (CNS) A4 (210X 297mm) binding

536919 A7536919 A7

而且根據本發明之-種較佳實施例，處理電路提供存在一組至少三預定材料中之至少二材料之輸出指示。另根據本發明之一種較佳實施例，處理電路存至少三預定材料之輸出指示。，、存在再根據本發明之一種較佳實施例，處理電路提供存在銅之輸出指示。、另外根據本發明之一種較佳實施例，處理電路提供存在金之輸出指示。而且根據本發明之一種較佳實施例，處理電路提供存在一種焊料掩模材料之輸出指示。另根據本發明之一種較佳實施例，處理電路提供存在銀之輸出指示。再根據本發明之一種較佳實施例，處理電路提供存在光敏抗蝕劑殘留物之輸出指示。另外根據本發明之一種較佳實施例，處理電路提供存在氧化金屬之輸出指示。而且根據本發明之一種較佳實施例，影像輸出表示如通過半透明表層所見電路之一部份。根據本發明之另一較佳實施例，也提供一種自動化光學叔查裝置，適合檢查有圖案物品，包括至少一檢測器，通過一至少部份覆蓋有圖案物品之半透明表層，提供有圖案物品之至少一部份之影像輸出，及處理電路，接收影像 ___ -34- ΐ紙張尺度適用巾@ @家標準(CNS) Μ驗(^^97公爱） ψ 裝訂參 32五、發明説明（輸出，並提供一另根據本發明路。再根據本發明板0 獨互於半透明表層之輸出指示。路义一種較佳實施例，有圖案物品包括 < 一種較佳實施例，電路包括印刷電另外根據本發明之_種較佳實施例，印刷電球狀格柵陣列基片。 I括 0而且根據本發明之-種較佳實施例，處理電路可操作半2表層所覆蓋之圖案之諸部份及半透明表層所未復盖 < 圖案 < 諸部份。根據本發明之另-較佳實施例，也提供裝置，供在彩色影像’纟-與二彩色粒子數之第_者關聯之第一 / 位與一與二彩色粒子數之第二者關聯之第二彩色部位之間’確定邊界之地點’影像包含至少二彩色粒子數，第一彩色邵位及第二彩色部位均包含在彩色影像中，該裝置包括邊界識別單元，可操作在第—彩色部位與第二^色部= 之間識別近似邊界地點，一候選邊界確定單元，可操作在第-彩色#位及第二彩“位之間確定許多候選邊界地點’許多候選邊界地點各係藉應用選自許多邊界地點方法當中之一對應邊界地點方法所確定，一方法選擇器‘’，可操作自許多邊界地點方法當中選擇—方法作為一種較佳方法，及邊界確定單元，可操作指定與較佳方法關聯之許多 -35- 本紙張尺度適用中國國家標準(CNS) Α4規格(210X 297公釐） 536919Furthermore, according to a preferred embodiment of the present invention, the processing circuit provides an output indication of at least two materials stored in a set of at least three predetermined materials. According to a preferred embodiment of the present invention, the processing circuit stores output indications of at least three predetermined materials. According to a preferred embodiment of the present invention, the processing circuit provides an output indication of the presence of copper. In addition, according to a preferred embodiment of the present invention, the processing circuit provides an output indication of the existence of gold. Furthermore, according to a preferred embodiment of the present invention, the processing circuit provides an output indication in the presence of a solder mask material. According to a preferred embodiment of the present invention, the processing circuit provides an output indication that silver is present. According to a preferred embodiment of the present invention, the processing circuit provides an output indication of the presence of a photoresist residue. In addition, according to a preferred embodiment of the present invention, the processing circuit provides an output indication of the presence of metal oxide. Furthermore, according to a preferred embodiment of the present invention, the image output represents a portion of the circuit as seen through a translucent surface layer. According to another preferred embodiment of the present invention, an automated optical tertiary inspection device is also provided, suitable for inspecting patterned articles, including at least one detector, and providing patterned articles through a translucent surface layer at least partially covered with patterned articles. At least a part of the image output, and processing circuit, receiving the image ___ -34- ΐ Paper size applicable towel @ @ 家标准 (CNS) Ｍ测 (^^ 97 公爱) ψ Binding reference 32 V. Description of the invention ( Output and provide another way according to the present invention. According to the present invention, the board 0 is independent of the output indication of the translucent surface layer. Lu Yi A preferred embodiment, the patterned article includes < a preferred embodiment, the circuit includes printing In addition, according to a preferred embodiment of the present invention, an electric spherical grid array substrate is printed. I includes 0 and according to a preferred embodiment of the present invention, the processing circuit can operate half of the patterns covered by the surface layer. Parts and parts of the translucent surface layer are not covered with the <patterns> According to another-preferred embodiment of the present invention, a device is also provided for the color image '纟-and the number of two-color particles_ The image of the 'determined boundary location' between the associated first / bit and the second colored part associated with the second one with the second colored particle number contains at least two colored particle numbers. Contained in a color image, the device includes a boundary recognition unit operable to identify an approximate boundary location between the first-color part and the second color part =, a candidate boundary determination unit operable at the -color # position and the Two-color "determine many candidate boundary locations between bits' Many candidate boundary locations are determined by applying one of many boundary location methods corresponding to the boundary location method, a method selector", operable from many boundary location methods Among them, the method is a better method, and the boundary determination unit is operable to designate many associated with the better method. -35- This paper size applies to China National Standard (CNS) A4 specification (210X 297 mm) 536919

候選邊界地點之一作a邊叉访 . 作為逯界，藉以在第一彩色部位及第二彩色部位之間確定邊界之地點。根據本發明之再一較佳實施％，也提供$置，供在彩色影像’在-與二彩色粒子數之第一者關聯之第一彩色部位與-與二彩色粒子數之第二者關聯之第二彩色部位之間，較邊界之地點，影像包含至少：彩色粒子數，第— 心色邛位及第一彩色部位均包含在彩色影像中，該裝置包括邊界分析圖單元，可操作提供邊界分析圖，供每一對性貝不同色t子數’包括供在彩色粒子數之間識別邊界地點之-種Μ方法之指示，輕識別單&可操作在第一彩色部位及第二彩色部位之間識料似邊界地點，及邊界確定單元可操作使用指示之較佳方法，在第一彩色部位與第二彩色部位之間，在邊界分析圖確定邊界地點，供在第一彩色粒子數與第二彩色粒子數之間識別邊界地點。根據本發明之又一較佳實施例，也提供裝置，供自彩色影像產生一彩色形態圖，該裝置包括一縮小單元，可操作鈿小彩色影像，以產生一包括許多彩色粒子數之縮小影像，邊界確定單元，可操作在許多彩色粒子數之相鄰者之間，在鈿小影像確定邊界，一分段單元，可操作將縮小影像分段，以產生一包括邊界及至少一非邊界部位之二進制影像’一骨架產生單元，可操作產生至少一非邊界部位之骨架景>像，骨架影像包括許多部份，及_指定單元，可操 -36- 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐)One of the candidate boundary locations is a side-border interview. As a boundary, the location of the boundary is determined between the first colored part and the second colored part. According to yet another preferred embodiment of the present invention, it is also provided for the color image to be associated with the first color part associated with the first number of two colored particles and the second number particle with the second number of colored particles. Between the second colored part and the boundary part, the image contains at least: the number of colored particles, the first-colored position and the first colored part are included in the color image. The device includes a boundary analysis map unit, which can be operated and provided. Boundary analysis chart, for each pair of different colors t number of sub-frames' Includes an indication of the M method for identifying the boundary location between the number of colored particles, light identification sheet & can be operated on the first color part and second The boundary locations between the colored parts are identified, and the boundary determination unit is operable to use a better method of indicating. Between the first and second colored parts, the boundary analysis map is used to determine the boundary positions for the first colored particles. The boundary point is identified between the number and the number of the second colored particles. According to still another preferred embodiment of the present invention, a device is also provided for generating a color morphology map from a color image. The device includes a reduction unit that can operate a small color image to generate a reduced image including a number of colored particles. , Boundary determination unit, which can be operated between the neighbors of many colored particles, to determine the boundary in the small image, a segmentation unit, which can be operated to reduce the image segmentation to produce a boundary and at least one non-boundary part The binary image 'a skeleton generating unit can be operated to generate at least one skeleton scene of a non-boundary part> image, the skeleton image includes many parts, and the designated unit can be manipulated ) A4 size (210X 297mm)

裝訂參 34 五、發明説明（作指定一對應於彩色一… 部份。子數 < 一义彩色識別至骨架之每一像識別在=月又另一較佳實施例，也提供裝置，供在影 r . 滑曲線之不規律，該裝置包括-識別單元，可操作沿平滑曲線，則古产、了少-相鄭點在有;許多點，有序之許多點各有至有序之許多點之每—二法線方向單元，可操作在 ^ 母點確足至平滑曲線之法線方向，及一不規律早70 ’可操作至少部份依據局部法線差，包括在該點至平滑曲線之法線方向與在該點之至少一相鄰點至平滑曲線之法線方向間之差’識別有序之許多點之一點為不規律之一部份。根據本發明之再_較佳實施例，也提供裝置，供將在物體〈影像之不規律分類，該裝置包括-尺寸確定單元，可操作確定一供不規律之特色深度及特色孔徑，及一分類單元可操作至少邵份依據尺寸確定單元之輸出將諸特色分類。根據本發明之又一較佳實施例，也提供裝置，供在物體 < 影像之一部份識別缺陷，該部份靠近一在影像識別為邊緣之地點’該裝置包括一除外方向單元，可操作識別一關聯與邊緣之除外方向，及缺陷單元，可操作靠近識別邊緣之地點’在不包括除外方向之許多方向搜尋缺陷。根據本發明之另一較佳實施例，也提供裝置，供在物 -37- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) ^36919 A7Binding parameters 34 V. Description of the invention (specify one corresponds to the color one ... part. Sub-number < one color identification to each image of the skeleton identification in another month, another preferred embodiment, also provides a device for In the shadow r. The irregularity of the sliding curve, the device includes a recognition unit, which can be operated along the smooth curve, then there are ancient and lesser-phase Zheng points; there are many points, and there are many points in order. Many points of each-two normal direction unit, can be operated in the direction from the ^ mother point to the normal direction of the smooth curve, and an irregular as early as 70 'can be operated at least partly based on the local normal difference, including at this point to The difference between the normal direction of the smooth curve and at least one adjacent point to the normal direction of the smooth curve 'identifies many points that are ordered as an irregular part. According to the present invention, the comparison The preferred embodiment also provides a device for classifying the irregularity of the image in the object. The device includes a size determination unit operable to determine a characteristic depth and characteristic aperture for the irregularity, and a classification unit operable at least Determine unit based on size The output classifies the features. According to yet another preferred embodiment of the present invention, a device is also provided for identifying defects on a part of the object < image, the part being close to a place where the image is identified as an edge ' Includes an excluded direction unit, operable to identify an excluded direction associated with an edge, and a defective unit operable to identify a location near the identified edge 'to search for defects in many directions excluding the excluded direction. According to another preferred embodiment of the present invention ， A device is also provided for the object-37- This paper size applies Chinese National Standard (CNS) A4 (210 X 297 mm) ^ 36919 A7

體之影像之一部位識为丨I ^ 成另】表面缺陷，影像包括很多像素，該 I置包括-子取樣單元’可操作根據—子取樣比對該部位子取’因此生產琢部位之子取樣影像，缺陷識別單元，可操作在子取樣影像識別表面缺陷，及一對應單元，可操作在對應於在子取樣影像之表面缺陷之部位識別表面缺陷。人根據本發明之再一較佳實施你]，也提供一種纟法，供檢查有圖案物品，包括提供有許多元件之有圖案物品之至 J # ^刀之夕色影像輸出，並接收多色影像輸出，以及提七、至/下列檢查功能之一：在許多元件之至少一元件檢查金屬塗層，及在許多元件之至少一元件檢查至少部份透明塗層。根據本發明之又一較佳實施例，也提供一種方法，供檢查有圖案物品，包括提供包括許多元件之有圖案物品之至少一邵份之多色影像輸出，及接收多色影像輸出，並提供下列禾X查功忐·在許多元件檢查在至少一元件之金屬塗層及在許多元件檢查在至少一元件之至少部份透明塗層。根據本發明之另一較佳實施例，也提供一種方法，供叔查物品，包括提供物品之至少一部份之多色影像輸出，及接收多色影像輸出，並提供邊緣檢測至子像素準確度，可操作辨別不同彩色之部位當中之邊界。One part of the image of the body is identified as “I ^”, the surface defect, the image includes a lot of pixels, the I set includes-sub-sampling unit 'operable according to-sub-sampling ratio to take the part', so the sub-sampling of the production part The image and defect recognition unit is operable to identify a surface defect in a sub-sampling image, and a corresponding unit is operable to identify a surface defect in a portion corresponding to the surface defect in the sub-sampling image. People according to yet another preferred embodiment of the present invention], and also provides a method for inspecting patterned items, including patterned items provided with many elements up to J # ^ Sword color image output, and receiving multi-color Image output, and one of the following inspection functions: inspection of metal coatings on at least one of many components, and inspection of at least partially clear coatings on at least one of many components. According to yet another preferred embodiment of the present invention, there is also provided a method for inspecting a patterned article, including providing at least one multi-color image output of a patterned article including a plurality of elements, and receiving the multi-color image output, and The following are provided: Check the metal coating on at least one component on many components and the at least partially transparent coating on at least one component on many components. According to another preferred embodiment of the present invention, a method is also provided for unchecking articles, including providing multi-color image output of at least a part of the article, and receiving multi-color image output, and providing edge detection to sub-pixel accuracy Degrees, can be used to identify the boundaries among different colored parts.

裝 -38-Equipment -38-

/據本發明之再-較佳實施例，查Y物=，包括提供一有至少三不同部位之物品之至少 /伤又以像輸出，該等部位各至少藉一光學特徵予以孕。”並接收衫像輸出，並提供邊緣檢測至子像素準確度可操作辨別至少三不同部位之至少二部位之邊界。 :據本發明之又一較佳實施例，也提供一種方法，^ 叔且物品，包括提供物品/ According to the re-preferred embodiment of the present invention, checking Y objects = including providing at least three different parts of the object at least / wound and output as an image, each of these parts is pregnant by at least one optical feature. "And receive shirt image output, and provide edge detection to sub-pixel accuracy to operate to identify the boundary of at least two different parts of at least three different parts .: According to yet another preferred embodiment of the present invention, a method is also provided. Items, including supplies

裝收影像輸出，並至少*依〜像輪出，《Install the image output, and at least * like ~ out,

依據在有共同邊界之物品之I 根二同部位識別材料之影像輸出提供剛-。 :據本發明之另一較佳實施例，也提供】： ^旦物品’包括提供物品之至少—部份之及接收多㊂与德私山巴心像輸出：色，二:像如出，並提供至少下列檢查功能之訂Provide rigid-based on the image output of the I-identical material at the same location of the item with a common boundary. : According to another preferred embodiment of the present invention, it is also provided]: ^ Once the item 'includes at least-part of the provided item and the reception of Duo and Deshen Shanpa heart image output: color, two: like the image, and Provides at least the following inspection functions

K刀奴為-包括許多彩色部位，各有許〆可能性之-之分段彩色影像，〔表性’ 為-包括數部位及邊界之二進制影像，該等;象w :，以提供一表示諸部位之骨架，以及指定：：之形態, 架中之諸元件。 y色邵位至，根據本發明之再一較佳實施例，檢查物品，包括提供物品之至少 :~種方法，供及接收多色影像輸出’並提供表示物：：=影像輸出，影: 象包括物品之至少-部份之形態腐：多色4像’彩' '據本發明之又-較佳實施例，也提供—種、。種万法，供 -39 - 本紙張尺紐^^^^1^^2謂97公 17 536919 A7 B7 五、發明説明（） 37 檢查物品，包括提供物品之至少一部份之多色影像輸出，及接收多色影像輸出，並提供表示物品之彩色影像，彩色影像包括一在物品之不同部位之間指示邊界，並在每一有共同邊界之不同部位識別材料之圖。K knife slave is-a segmented color image that includes many colored parts, each of which has a possibility of being different, [representative 'is-a binary image that includes several parts and boundaries, etc .; like w: to provide a representation The skeleton of various parts, and the designation :: the form, the components in the frame. According to yet another preferred embodiment of the present invention, inspecting articles includes providing at least: ~ methods for supplying and receiving multi-color image outputs' and providing representations: = image outputs, shadows: The image includes at least-part of the form of the rot: multicolor 4 like 'color' According to yet another preferred embodiment of the present invention, a species is also provided. Various methods for -39-This paper ruler ^^^^ 1 ^^ 2 is 97 public 17 536919 A7 B7 V. Description of the invention () 37 Inspection items, including providing multi-color image output of at least part of the items , And receive multi-color image output, and provide a color image representing the article, the color image includes a map indicating the boundary between different parts of the article, and identifying the material at each different part with a common boundary.

根據本發明之另一較佳實施例，也提供一種方法，供檢查有圖案物品，包括提供包括至少三材料，有可檢測邊緣之有圖案物品之至少一部份之影像輸出，及接收影像輸出，並提供檢查功能，包括影像輸出之二進制化，以提供一包括部位影像資料及邊界影像資料之二進制影像。根據本發明之再一較佳實施例，也提供一種方法，供檢查有圖案物品，包括提供包括許多材料，在許多材料之間有可檢測邊界之有圖案物品之至少一部份之影像輸出，及接收影像輸出，並提供檢查功能，包括關於邊界影像資料之圖案分析，以提供沿邊界存在凹隙及凸起部之指示。 Φ 根據本發明之又一較佳實施例，也提供一種方法，供檢查物品，包括提供一包括多材料，在材料之間有可檢測邊界之物品之至少一部份之影像輸出，接收影像輸出，並提供至少下列檢查功能之一：影像輸出之分段為一包括部位影像資料及邊界影像資料之分段影像，使用一第一技術分析該部位影像資料，及使用一不同於第一技術之第二技術分析邊界影像資料。根據本發明之另一較佳實施例，也提供一種方法，供 -40- 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐） 536919 A7 B7 五、發明説明（） 38 檢查物品，包括提供一包括多材料，在材料之間有可檢測邊界之物品之至少一部份之多色影像輸出，及接收多色影像輸出，並提供至少下列檢查功能之一：多色影像輸出之分段為一包括選自一組預定彩色當中之彩色部位之分段彩色影像，使用一第一技術分析關於第一彩色之部位影像資料，及使用一不同於第一技術之第二技術分析關於第二彩色之部位影像資料。根據本發明之再一較佳實施例，也提供一種方法，供檢查物品，包括提供一包括多材料，在材料之間有可檢測邊界之物品之至少一部份之影像輸出，接收影像輸出，並提供至少下列檢查功能之一：影像輸出之分段為一包括部位影像資料及邊界影像資料之分段影像，並以一種不受至少有些邊界影像資料影響之方式，使部位影像資料平滑。根據本發明之又一較佳實施例，也提供一種方法，供檢查電路，包括提供包括多材料之電路之至少一部份之影像輸出，及接收影像輸出，並提供存在至少一預定材料之輸出指示。根據本發明之另一較佳實施例，也提供一種方法，供檢查有圖案物品，包括通過一至少部份覆蓋有圖案物品之半透明表層，提供有圖案物品之至少一部份之影像輸出，及接收影像輸出，並提供一獨立於半透明表層之輸出指示。 -41 - 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐)According to another preferred embodiment of the present invention, there is also provided a method for inspecting a patterned article, including providing an image output including at least three materials and at least a portion of a patterned article having a detectable edge, and receiving the image output It also provides inspection functions, including binary image output, to provide a binary image including part image data and boundary image data. According to yet another preferred embodiment of the present invention, there is also provided a method for inspecting a patterned article, including providing an image output of at least a portion of a patterned article including a plurality of materials with a detectable boundary between the plurality of materials, And receive image output, and provide inspection functions, including pattern analysis of boundary image data, to provide an indication of the existence of gaps and protrusions along the boundary. Φ According to yet another preferred embodiment of the present invention, a method is also provided for inspecting an article, including providing an image output of at least a portion of an article including multiple materials with a detectable boundary between the materials, and receiving the image output And provide at least one of the following inspection functions: the segmentation of the image output is a segmented image including part image data and boundary image data, using a first technique to analyze the part image data, and using a different technique than the first technique The second technique analyzes the boundary image data. According to another preferred embodiment of the present invention, there is also provided a method for -40- this paper size is applicable to Chinese National Standard (CNS) A4 (210X 297 mm) 536919 A7 B7 V. Description of the invention , Including providing a multi-color image output including at least a part of an item that includes multiple materials with a detectable boundary between the materials, and receiving the multi-color image output, and providing at least one of the following inspection functions: Segmentation is a segmented color image including color parts selected from a set of predetermined colors, using a first technique to analyze the image data of the part in the first color, and using a second technique different from the first technique to analyze the Image data of the second color. According to yet another preferred embodiment of the present invention, a method is also provided for inspecting an article, including providing an image output of at least a portion of an article including multiple materials with a detectable boundary between the materials, and receiving the image output, It also provides at least one of the following inspection functions: the segmentation of the image output is a segmented image including part image data and boundary image data, and the part image data is smoothed in a manner not affected by at least some of the boundary image data. According to yet another preferred embodiment of the present invention, a method is also provided for inspecting a circuit, including providing an image output of at least a portion of a circuit including multiple materials, and receiving the image output, and providing an output in which at least one predetermined material is present. Instructions. According to another preferred embodiment of the present invention, there is also provided a method for inspecting a patterned article, including providing an image output of at least a portion of the patterned article through a translucent surface layer at least partially covered with the patterned article, And receive image output, and provide an output indication independent of the translucent surface layer. -41-This paper size applies to China National Standard (CNS) A4 (210 X 297 mm)

裝訂Binding

536919 A7 B7 五、發明説明（） 39 根據本發明之再一較佳實施例，也提供一種方法，供自動光學檢查物體之彩色影像，包括獲得物體之多色影像，處理多色影像及輸出一報告，指示候選缺陷及選自預一定組特色當中之特色，並提供一包圍每一候選缺陷及特色之部位之影像。根據本發明之又一較佳實施例，也提供一種裝置，供自動光學檢查物體之彩色影像，包括一感測器，可操作獲得物體之多色影像，一影像處理器，可操作處理多色影像並輸出一報告，指示候選缺陷及選自一預定組特色當中之特色，並提供一包圍每一候選缺陷及特色之部位之影像。另根據本發明之一種較佳實施例，處理器包括一二進制影像處理器，可操作處理自多色影像所產生之二進制影像，及一彩色影像處理器，可操作處理多色影像。再根據本發明之一種較佳實施例，該裝置也包括一形態影像處理器，可操作自一自多色影像導得之多彩色影像產生一骨架圖。附圖之簡要說明自下列詳細說明，配合附圖，將會更完全瞭解及察知本發明，在附圖中：圖1為一根據本發明之一種較佳實施例所構造及操作之影像分析系統之簡化方塊圖示；圖2為圖1之材料識別子單元170，其一種較佳實施之 -42- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) 536919536919 A7 B7 V. Description of the Invention () 39 According to yet another preferred embodiment of the present invention, a method is also provided for automatically optically inspecting a color image of an object, including obtaining a multi-color image of the object, processing the multi-color image, and outputting one. The report indicates candidate defects and features selected from a predetermined set of features, and provides an image of the area surrounding each candidate defect and feature. According to yet another preferred embodiment of the present invention, a device is also provided for automatically optically inspecting a color image of an object, including a sensor operable to obtain a multicolor image of the object, and an image processor operable to process the multicolor The image outputs a report indicating candidate defects and features selected from a predetermined set of features, and provides an image surrounding a portion of each candidate defect and feature. According to a preferred embodiment of the present invention, the processor includes a binary image processor operable to process binary images generated from multi-color images, and a color image processor operable to process multi-color images. According to a preferred embodiment of the present invention, the device also includes a morphological image processor, which is operable to generate a skeleton image from a multi-color image derived from the multi-color image. Brief Description of the Drawings The following detailed description, together with the drawings, will fully understand and perceive the present invention. In the drawings: FIG. 1 is an image analysis system constructed and operated according to a preferred embodiment of the present invention Figure 2 is a simplified block diagram; Figure 2 is the material identification sub-unit 170 of Figure 1, a preferred implementation of which is -42- This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) 536919

五、發明説明（ 40 簡化方塊圖示；圖3為圖2之裝置之部份，其一種較佳方法之簡化流程圖示；圖4為操作圖2之部位擴展單元173，其一種較佳方法又間化流程圖示；圖5及6分別為一較佳5χ5大核心及一較佳3χ3對應小核心之簡化晝面圖示；圖7為圖2之輪廓化單元ρ5，其一種較佳實施之簡化方塊圖示；圖8為操作圖7之最佳輪廓選擇器178，其一種較佳方法之間化流程圖示；圖9 11為使用於瞭解本發明之較佳cel構造及術語表之簡化畫面圖示；圖12為圖i之彩色表面缺陷子單元19〇，其一種較佳實施簡化方塊圖示；圖13為操作圖12之彩色w彩色2造成單元4〇〇，其一種較佳方法之簡化流程圖示；圖Η為圖12之方向允許影像造成器4〇2之第一部份，其一種較佳實施之簡化方塊圖示；圖15為一使用於瞭解圖14之裝置之操作之影像像素及相鄰虛擬cel像素之簡化晝面圖示；圖16為操作圖15之〇kdir圖計算單元416，其—種較 -43-V. Description of the invention (40 simplified block diagram; FIG. 3 is a part of the device of FIG. 2 and a simplified flow chart of a preferred method thereof; FIG. 4 is an operation of the location expansion unit 173 of FIG. 2 Figures 5 and 6 are simplified day-to-day diagrams of a preferred 5x5 large core and a preferred 3x3 corresponding small core, respectively; Figure 7 is a contouring unit ρ5 of Figure 2, which is a preferred implementation Figure 8 is a simplified block diagram; Figure 8 is a diagram illustrating a preferred method of interleaving the optimal contour selector 178 of Figure 7; Figures 9 and 11 are diagrams of the preferred cel structure and glossary for understanding the present invention Simplified screen illustration; Figure 12 is a simplified block diagram of a color surface defect sub-unit 19 of FIG. I, which is a preferred implementation; A simplified flow chart of the method; Figure Η is the first part of the direction allowing image generator 402 in Figure 12, a simplified block diagram of a preferred implementation thereof; Figure 15 is a device used to understand the device of Figure 14 Simplified day view of the operating image pixels and adjacent virtual cel pixels ; FIG. 16 is the operation of 〇kdir calculation unit 416 in FIG. 15, which - more kinds -43-

536919 A7 B7 五、發明説明（） 41 佳方法之簡化流程圖示；圖 17 為圖16之步騾418 ，其一種較佳實施之簡化流程圖示圖 18 為圖16之步騾420 ，其一種較佳實施之簡化流程圖，圖 19 為使用於暸解圖18之方法之腐蝕過程，其一實例之簡化畫面圖示；圖 20 為一使用於暸解圖 18之方法之表，示 okdir 值，對應測量方向，及一供測量之較佳梯度運算符；圖 21 為圖12之方向允許影像造成器402之第二部份，其一種較佳操作方法之簡化流程圖TF，圖 22 使用於瞭解圖21 之方法之預定像素，其一種較佳選擇之簡化晝面圖示；圖 23A為一表，例示供佼 .用在圖21之步騾462之較佳試探圖 23B為簡化畫面圖示，示一供使用於暸解圖 23A 之3 X3虛擬展開之9像素之命名協定；圖 24 為操作圖12之裝置之一部份，其一種較佳方法之簡化流程圖示；圖 25 為圖24之步驟464 ，其一種較佳實施之簡化流程圖示 j 圖 26 為圖25之步驟470 ，其一種較佳實施之簡化流程 -44- 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐) 536919 A7 B7 五、發明説明（） 42 圖示圖 27為操作圖1之線及空間骨架子單元 230 之一種較佳方法之化流程圖示；圖 28A及28B為使用於暸解圖27之方法之一陣列像素在部份腐蝕前及後之簡化畫面圖示；圖 29使用於暸解圖27之方法之Euler狀況之簡化晝面圖示 j 圖 30A —30C為使用於瞭解操作圖1之形態子單元 240 之一種較佳方法之簡化畫面圖示；圖 31A —31C為使用於暸解圖30A —30C之運算符之簡化畫面圖 7JT，圖 32為彩色形態分析影像之一種較佳方法之簡化流程圖示圖 33A及33B，合在一起，包含操作圖 1之凹隙及凸起部子單元220之一種較佳方法之簡化流程圖示，圖 34為有助於暸解圖33A之步驟890 之簡化晝面圖示圖 35為一予以分析之影像之一部份之簡化畫面圖示，可用於暸解圖33A及33B之方法；圖 36為使用於暸解圖33B之方法步驟950之 — 陣列虛擬 cel像素之簡化畫面圖示；圖 37A及37B為輪廓之實例之簡化畫面圖不， -45- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) 536919 A7 B7 五、發明説明（） 43 圖38為一凸起部之代表性實例之簡化畫面圖示；圖39A及39B為使用於暸解圖33B之方法步騾980之一陣列虛擬cel像素之簡化畫面圖示；圖39C及39D為配合圖1之系統使用之方法之簡化流程圖示；以及裝圖 40為可用於暸解本發明之操作之電路，其一實例之簡化畫面圖示。較佳實施例之詳細說明概要536919 A7 B7 V. Description of the invention () 41 Simplified flow diagram of the best method; Figure 17 is a simplified flow diagram of step 418 of FIG. 16, which is a preferred implementation. Simplified flowchart of the preferred implementation, FIG. 19 is a simplified screen diagram of an example used to understand the corrosion process of the method of FIG. 18; FIG. 20 is a table used to understand the method of FIG. 18, showing okdir values, corresponding Measurement direction and a better gradient operator for measurement; Figure 21 is the second part of the direction allowable image generator 402 of Figure 12, a simplified flowchart TF of a preferred method of operation, Figure 22 is used to understand the diagram The predetermined pixel of the 21 method is a simplified daytime diagram of a preferred choice; FIG. 23A is a table illustrating the best examples. The better test used in step 462 of FIG. 21 is shown in FIG. 23B as a simplified screen diagram. A naming convention for understanding the 9-pixel 3 X3 virtual expansion of FIG. 23A; FIG. 24 is a part of the device for operating FIG. 12, and a preferred method thereof Simplified process diagram; Figure 25 is step 464 of FIG. 24, a simplified process diagram of a preferred implementation. Figure 26 is step 470 of FIG. 25, a simplified process of a preferred implementation. National Standard (CNS) A4 specification (210X 297 mm) 536919 A7 B7 V. Description of the invention () 42 Figure Figure 27 is a diagram of a preferred method of operating the line and space skeleton subunit 230 of Figure 1; FIGS. 28A and 28B are simplified screen diagrams of array pixels before and after partial corrosion, which is used to understand the method of FIG. 27; FIG. 29 is a simplified day-time diagram of Euler conditions used to understand the method of FIG. 27; FIG. 30A —30C is a simplified screen diagram used to understand a better method of operating the morphological subunit 240 of FIG. 1; FIG. 31A—31C is a simplified screen diagram used to understand the operators of FIG. 30A—30C FIG. 7JT, FIG. 32 is color Simplified flow diagram of a better method for morphological analysis Figures 33A and 33B, taken together, include operations Simplified flow diagram of a preferred method of the recessed and raised sub-unit 220 of FIG. 1, FIG. 34 is a simplified day-time diagram useful for understanding step 890 of FIG. 33A, and FIG. 35 is an image to be analyzed Part of the simplified screen diagram can be used to understand the method of Figures 33A and 33B; Figure 36 is a simplified screen diagram of the array of virtual cel pixels used to understand the method step 950 of Figure 33B; Figures 37A and 37B are outlines The simplified picture of the example is not, -45- This paper size is applicable to Chinese National Standard (CNS) A4 specification (210 X 297 mm) 536919 A7 B7 V. Description of the invention () 43 Figure 38 is a representative of a raised part Simplified screen diagram of the example; Figures 39A and 39B are simplified screen diagrams of an array of virtual cel pixels used to understand the method step 980 of Figure 33B; Figures 39C and 39D are simplified processes used in conjunction with the system of Figure 1 Figures; and Figure 40 is a simplified screen diagram of an example of a circuit that can be used to understand the operation of the present invention. Detailed description of the preferred embodiment

現請參照圖1，其為一根據本發明之一種較佳實施例所 .構造及操作之影像分析系統100之簡化方塊圖示。請予察知，雖然圖1之系統特別使用於分析一表示有圖案物體，例如BGA或其他電路之影像，但圖1之系統通常可使用於影像分析，不論予以分析之影像是否表示有圖案物體或任何其他物體。雖然在本案說明書係參照BGA供例示本發明之目的，但使用於本案說明書之BGA —詞，將視為指並另包括印刷電路板基片’層壓印刷電路板’引線框架，平板顯示器，混合式晶片封裝基片，帶自動化接合基片，及任何適當有圖案物體，包括如可使用於醫療移植物之各不同蝕刻及壓花金屬基片。影像分析系統1〇〇較佳為予以包括在一檢查系統，包含照明，影像獲取，物體運送，及軟體基礎之影像處理子系 -46- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐） 536919 A7 B7 五、發明説明（） 44 統。因此，請予察知，在影像分析系統100予以分析之影像，包含物體，諸如一有圖案物體，如BGA之影像之情形，一般為採用一適當影像獲取系統（未示），以捕捉物體之影像，並且影像分析系統100予以分析之影像一般為以在此項技藝所熟知之適當方式，包括聚焦，補償，再取樣，及對準，予以校正。一適合提供影像至影像分析系統 100之適當影像獲取系統，如在此項技藝所熟知，一般為包含至少一提供多色影像輸出之檢測器。此種影像獲取系統之輸出，或其他適當裝置所產生之任何類似影像，本文中通稱為’’彩色影像’’。將行予以分析之輸入影像，如在此項技藝所熟知，一般為包含成一種適當彩色格式之彩色影像。較佳為，以 RGB (紅一綠一監）格式’在南位元解析度，諸如24 —位元解析度提供彩色影像。影像分析系統100較佳為可操作分析輸入影像，並傳遞分析之結果供另外處理，較佳為一軟體影像處理單元，或供藉由適當應用介面，直接報告至運算符。較佳為，如在此項技藝所熟知，影像分析系統100以實製硬體予以實施。雖然將行在通用目的電腦運作之影像分析系統100之軟體實施，據信為可能，並且據信將會完全可操作，但以硬體實施為較佳，俾達成較快性能。為容易說明本發明，該方法通常予以說明，使能以硬體或軟體實 -47- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐）Please refer to FIG. 1, which is a simplified block diagram of an image analysis system 100 constructed and operated according to a preferred embodiment of the present invention. Please note that although the system of Figure 1 is specifically used to analyze an image representing a patterned object, such as a BGA or other circuit, the system of Figure 1 can usually be used for image analysis, regardless of whether the image being analyzed indicates a patterned object or Any other object. Although the description in this case refers to the BGA for illustrating the purpose of the present invention, the word BGA used in the description of this case will be deemed to refer to and additionally include the printed circuit board substrate 'laminated printed circuit board' lead frame, flat panel display, hybrid Wafer packaging substrates, substrates with automated bonding, and any suitable patterned objects, including, for example, various etched and embossed metal substrates that can be used in medical implants. The image analysis system 100 is preferably included in an inspection system, which includes lighting, image acquisition, object transportation, and software-based image processing subsystems-46- This paper standard applies to China National Standard (CNS) A4 specifications (210X297 (Mm) 536919 A7 B7 V. Description of the invention () 44 system. Therefore, please be aware that in the case where the image analyzed by the image analysis system 100 includes an object, such as a patterned object, such as a BGA image, an appropriate image acquisition system (not shown) is generally used to capture the image of the object The images analyzed by the image analysis system 100 are generally corrected in a suitable manner known in the art, including focusing, compensation, resampling, and alignment. A suitable image acquisition system suitable for providing images to the image analysis system 100, as is well known in the art, generally includes at least one detector that provides multi-color image output. The output of such an image acquisition system, or any similar image produced by other suitable devices, is referred to herein as '' color image ''. The input image to be analyzed is, as is well known in the art, a color image that is contained in a suitable color format. Preferably, a color image is provided in a RGB (red-green-one-monitor) format at a South bit resolution, such as a 24-bit resolution. The image analysis system 100 is preferably operable to analyze the input image and pass the results of the analysis for additional processing, preferably a software image processing unit, or for reporting directly to the operator through an appropriate application interface. Preferably, as is well known in the art, the image analysis system 100 is implemented in actual hardware. Although the software implementation of the image analysis system 100 operating on a general-purpose computer is believed to be possible and it is believed to be fully operational, it is better to implement it in hardware to achieve faster performance. In order to easily explain the present invention, the method is usually explained, and can be implemented in hardware or software.

裝 536919 A7 B7 五、發明説明（_ ) 45 施本發明之原理。影像分析系統 100較佳為包含一彩色處理器，或 ”CLRP”單元120，其較佳為接收關於輸入影像之彩色資訊，並且如以下所另外說明，較佳為可操作依據所接收之彩色資訊分析輸入影像。影像分析系統1〇〇也較佳為包含一 CABS單元130，其較佳為接收關於輸入影像之單色（也稱作灰度級資訊），並且較佳為如以下所另外說明，可操作依據所接收之灰度級資訊分析輸入影像。單色資訊可包含一彩色中之資訊，一般但不一定為紅色，並且一般為包含高位元數，諸如8 位元之灰度級資訊。以下參照圖33A及33B說明CABS 單元130之一種較佳實施。影像分析系統100也較佳為包含一 SKFD單元140，其較佳為接收一輸出作為輸入，以下所另外說明，並且一般為包含二進制化資訊，自CABS單元130，及彩色資訊自 CLRP單元120。SKFD單元140較佳為如以下所另外說明，可操作分析至其之輸入，並自其產生特徵資訊，一般為包括見於輸入影像之輸入影像及形態特色之骨架表示。影像分析系統100也較佳為包含一同步單元150。較佳為並為獲得影像分析系統100之較快性能，CLRP單元 120，CABS單元130，及SKFD單元140並行操作。同步單元150較佳為與每一 CLRP單元120，CABS單元130， -48- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐)Equipment 536919 A7 B7 V. Description of the invention (_) 45 The principle of the present invention is applied. The image analysis system 100 preferably includes a color processor, or a "CLRP" unit 120, which preferably receives color information about the input image, and as further explained below, it is preferably operable based on the received color information Analyze the input image. The image analysis system 100 also preferably includes a CABS unit 130, which preferably receives monochrome (also referred to as gray-scale information) about the input image, and is preferably as explained below, and can be operated based on The received gray level information analyzes the input image. Monochrome information can include information in a color, which is generally but not necessarily red, and generally includes a high number of bits, such as 8-bit grayscale information. A preferred implementation of the CABS unit 130 is described below with reference to FIGS. 33A and 33B. The image analysis system 100 also preferably includes an SKFD unit 140, which preferably receives an output as an input, as described below, and generally includes binary information from the CABS unit 130 and color information from the CLRP unit 120. The SKFD unit 140 is preferably operatively analyzed to its input, as described below, and generates feature information from it, typically a skeleton representation that includes the input image and morphological features found in the input image. The image analysis system 100 also preferably includes a synchronization unit 150. Preferably and in order to obtain faster performance of the image analysis system 100, the CLRP unit 120, the CABS unit 130, and the SKFD unit 140 operate in parallel. The synchronizing unit 150 is preferably associated with each CLRP unit 120, CABS unit 130, -48- The paper size is applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm)

裝訂Binding

536919 A7 B7 五、發明説明（） 46 及SKFD單元140操作關聯，並且較佳為自每一 CLRP單元120，CABS單元130，及SKFD單元140接收輸出。使用任何適當同步方法，諸如在此項技藝熟知之硬體基礎之延遲及同步方法，同步單元150較佳為在其所接收之每一輸入相互同步操作，並傳遞同步之輸入作為影像分析系統 100之輸出。較佳為影像分析系統100產生若干報告，其包括關於在一予以檢查之物體所檢測之各不同特色及缺陷之資訊。另外影像分析系統100產生每一特色及缺陷之快照影像，以及在物品上所考慮之特定部位之快照缺陷，如使用者在檢查前所預先界定。快照影像一般為一包圍每一特色，缺陷及所考慮之預先界定部位之部位之多色影像。較佳為，影像分析系統100之快照影像輸出，可藉一後處理子系統（未示）予以後處理，俾產生關於所檢查物體之結構及缺陷之適當最後報告。 CLRP單元120較佳為包含下列子單元： 1. 一材料識別子單元170，其較佳為接收彩色影像資料輸入至CLRP單元120，並較佳為可操作自其導得關於在物體中所包含，其輸入影像為影像之材料之資訊。以下參照圖2更詳細說明材料識別子單元之一種較佳實施。 2. —材料輪廓子單元180，其較佳為自材料識別子單元 170接收輸出，並較佳為可操作自其導得在輸入影像表示 -49- 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐) 裝訂536919 A7 B7 V. Description of the Invention () 46 and SKFD unit 140 are operationally related, and preferably receive output from each CLRP unit 120, CABS unit 130, and SKFD unit 140. Using any suitable synchronization method, such as the hardware-based delay and synchronization method well known in the art, the synchronization unit 150 preferably operates in synchronization with each other on each input it receives, and passes the synchronized inputs as the image analysis system 100 Its output. Preferably, the image analysis system 100 generates a number of reports including information about various features and defects detected in an object to be inspected. In addition, the image analysis system 100 generates a snapshot image of each feature and defect, and a snapshot defect of a specific part considered on the article, as previously defined by the user before inspection. Snapshot images are generally multi-colored images that enclose each feature, defect, and pre-defined location under consideration. Preferably, the snapshot image output of the image analysis system 100 may be post-processed by a post-processing subsystem (not shown) to generate an appropriate final report on the structure and defects of the inspected object. The CLRP unit 120 preferably includes the following sub-units: 1. A material identification sub-unit 170, which preferably receives color image data and inputs it to the CLRP unit 120, and is preferably operable to derive information about inclusion in the object, The input image is the information of the material of the image. A preferred implementation of the material identification subunit will be described in more detail with reference to FIG. 2 below. 2. —Material profile subunit 180, which preferably receives output from the material identification subunit 170, and is preferably operable to derive its output from the input image representation -49- This paper size applies to China National Standard (CNS) A4 specifications (210X 297mm) Staple

536919 A7 B7 五、發明説明（） 47 輪廓及/或邊界之輪廓資訊。以下也參照圖2更詳細說明材料輪廓子單元180之一種較佳實施。請予察知，在輸入影像不包含物體之影像之情形，如在整個本案說明書及申請專利範圍中特別但不專就材料識別子單元170及材料輪廓子單元180所使用”材料” 一詞，係用以在輸入影像之特色，其在分析時，如果影像表示一物體，見為對應於在一將會表示其影像之物體中之一種或多種材料之特色。換言之，以”材料’'一詞之擴大瞭解及定義，材料識別子單元170及材料輪廓子單元180所進行之操作，據信也可使用於分析一不包含物體之影像之影像。 3. —彩色表面缺陷子單元190，其較佳為接收彩色影像資料輸入至CLRP單元120，並較佳為可操作自輸入導得關於在輸入影像之表面缺陷之資訊，表面缺陷一般為對應於在一其輸入影像為影像之物體之缺陷。以下參照圖12 更詳細說明操作彩色表面缺陷子單元190之一種較佳方法。 CABS單元130較佳為包含下列子單元： 1. 一二進制化及cel產生子單元200，其較佳為接收單色影像資料輸入至CABS單元130，並較佳為依據單色影像資料，可操作自其導得表示對應於輸入影像之二進制影像之二進制資訊。另外，在本發明之一種較佳實施例，二 -50- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐)536919 A7 B7 V. Description of the invention () 47 Contour information of contour and / or boundary. A preferred implementation of the material contouring sub-unit 180 is also described in more detail below with reference to FIG. 2. Please be aware that in the case where the input image does not contain an image of an object, such as in the entire specification and patent application of this case, the term "material" is used specifically but not exclusively for the material identification sub-unit 170 and material contour sub-unit 180. For the characteristics of the input image, in the analysis, if the image represents an object, it is considered to be a feature corresponding to one or more materials in an object that will represent its image. In other words, with the expanded understanding and definition of the word "material", the operations performed by the material identification sub-unit 170 and the material contour sub-unit 180 are also believed to be useful for analyzing an image that does not contain an object. 3.-Color surface The defect subunit 190 preferably receives color image data and inputs it to the CLRP unit 120, and is preferably operable to self-input to obtain information about surface defects in the input image. The surface defects generally correspond to the input image in the input image. A defect of an image object. A better method of operating the color surface defect subunit 190 is described in more detail below with reference to FIG. 12. The CABS unit 130 preferably includes the following subunits: 1. A binarization and cel generation subunit 200, It is preferred to receive monochrome image data and input it to the CABS unit 130, and preferably based on the monochrome image data, operable to derive binary information representing a binary image corresponding to the input image. In addition, in one aspect of the present invention, Preferred embodiment, 2-50- This paper size applies to China National Standard (CNS) A4 (210 X 297 mm)

裝訂Binding

48 五、發明説明（ 2 =及Γ1產生子單元200自材料識別子單元170接收 :傻心彩色影像，並可操作自其導得表示對應於輸入縮二=進制影像之二進制資訊。二進制化及cel產生界早兀Λ也較佳為—4 3 —可操作界定在影像内之邊 I ’ “為藉輪廓元件，也稱作cel，並 :1—之說明之資訊。操作二進制化…生子單元200 <八所万去類似於以下參照圖7 ’供彩色影像之-組份所說明者。欠如在整個本案說明書及巾請專利範圍，特別是在說明 Μ’影像’或供處理資訊或處理影像之方法，所使用之進制"…在所有其形式，指有二狀態或二結果。在料例如二進制影像為影像之情形，其中包含在影像中 ^母一兀件’諸如像素’僅有二可能狀態，其可予以表示為Λ"及"白”，或則為，，0'，或”1”。在參照影像或影像資訊〈一進制一心因此係予使用對照於”灰階•，及.，彩色”諸闷：其ί该項技藝為熟知者，並且其指每-元件有超過1 可旎狀悲或結果之特定類型之影像。 2.凹：及凸起部單元22〇，其較佳為接收二進制化及 “丨產生單元200所輸出之⑷資訊，並較佳為可操作自 ” τ知關S在—進制化及eel產生單元綱所界定之邊之小邊界不規律’稱作凹隙及凸起部之資訊，並報告凹隙及凸起部作為輸出。以下參照圖33A133B更詳細說明 ___-51 - 本紙張尺度適财g S㈣準 536919 A7 B7 五、發明説明（） 49 操作凹隙及凸起部子單元220之一種較佳方法。 3. —灰小缺陷子單元210，其較佳為接收單色影像資料輸入至CABS單元130，並較佳為可操作使用單色影像資料識別在輸入影像之小缺陷，其在影像為物體之影像之情形，可指示在其影像為影像之物體上之缺陷。使用單色資訊識別小缺陷之方法，在此項技藝為熟知者，並例如說明於以上所提及，授予Aloni等人之美國專利5,586,058號，其揭示經予參考併入本案。 SKFD單元140較佳為包含下列子單元： 1. 一線及空間骨架子單元230，其較佳為自CABS單元 130接收二進制資訊輸出，一般為經由二進制化及cel產生子單元200，並較佳為可操作自其導得及產生作為說明輸入影像之輸出線及空間骨架資訊。一般為，線及骨架資訊係如在此項技藝所熟知，使用一種像素消耗方法所導得。以下參照圖27更詳細說明操作線及空間骨架子單元 230之一種較佳方法。 2. —形態子單元240，其較佳為接收線及空間骨架子單元230所輸出之線及空間骨架資訊，並較佳為可操作依據線及空間骨架資訊識別輸入影像之形態特色，並產生說明形態特色之資訊作為輸出。以下參照圖30A —30C更詳細說明操作形態子單元240之一種較佳方法。下列定義及概括討論，可有助於暸解本案說明書及申 -52- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐） 53691948 V. Description of the invention (2 = and Γ1 generating sub-unit 200 receives from the material identification sub-unit 170: silly color image, and can be operated from it to represent binary information corresponding to the input binary = binary image. Binarization and The cel generation boundary is also preferably -4 3-the edge defined in the image can be manipulated by the contour element, also known as cel, and: 1-the information of the description. Binarization of operations ... 200 < Basuowan is similar to the one described below with reference to Figure 7 'for color images-components. In the entire specification and patent scope of this case, especially in the description of the' image 'or for processing information or processing The method of image, the base "&...; in all its forms, refers to two states or two results. In the case of binary images such as images, which are contained in the image ^ mother and one element 'such as pixels' only There are two possible states, which can be expressed as Λ " and " white ", or, 0 ', or" 1 ". In the reference image or image information <unified by one base, therefore, it is used in contrast to" grey Tier •, and ., "Color": the technology is well-known, and it refers to a specific type of image with more than 1 sorrow or result per element. 2. Concave: and convex unit 22, which It is better to receive the binarization and "丨 information output from the generating unit 200, and it is preferably operable from" τ knows that the S-edge and the small boundary of the edge defined by the eel generating unit are irregular. ' It is called the information of the dents and bulges, and reports the dents and bulges as output. The following is a more detailed description with reference to Figure 33A133B ___- 51-This paper is suitable for gs㈣standard 536919 A7 B7 V. Description of the invention ( ) 49 A preferred method of operating the recessed and raised sub-unit 220. 3. —Gray small defect sub-unit 210, which preferably receives monochrome image data and inputs it to the CABS unit 130, and is preferably operable for use Monochrome image data identifies small defects in the input image. In the case where the image is an image of an object, it can indicate the defect on the object whose image is an image. The method of using the monochrome information to identify small defects in this technique is Familiar, as explained above It is mentioned that U.S. Patent No. 5,586,058, issued to Aloni et al., The disclosure of which is incorporated herein by reference. The SKFD unit 140 preferably includes the following subunits: 1. First line and space skeleton subunit 230, which is preferably a CABS unit 130 receives the binary information output, and generally generates the subunit 200 through binarization and cel, and is preferably operable to derive and generate the output line and spatial skeleton information as a description input image. Generally, the line and skeleton information system As is well known in the art, it is derived by using a pixel consumption method. A preferred method of operating the line and the space skeleton sub-unit 230 will be described in more detail below with reference to FIG. 27. 2. —morphology sub-unit 240, which preferably receives the line and space skeleton information output by the line and space skeleton sub-unit 230, and is preferably operable to identify the morphological characteristics of the input image based on the line and space skeleton information and generate Information describing morphological characteristics is used as output. Hereinafter, a preferred method of operating the subunit 240 will be described in more detail with reference to FIGS. 30A to 30C. The following definitions and general discussions can help to understand the specification and application of this case -52- This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) 536919

請專利範圍。 ^有大都為均勾彩色特徵之像素之粒子數，在本文稱 =大都為均勻彩色粒子數”。一般為，如在此項技藝所、將行予以分析之彩色影像，包含大都為均句彩色粒子數之許多部位。以 9 裝彩色格式或彩色空間，諸如例如：RGB格式，。口單獨之彩色值供紅，綠，及藍，以及服格式，包含單獨之值供色相，強度，及飽和，表示彩色影像，在此項技藝為熟知者。為說明簡單起見，在整個本案說明责及申請專^園，除非另外指日月’將使用RGB格式之實例’请予察知，在大多數情形，也可使用其他格式。訂Please patent scope. ^ The number of particles with pixels that are mostly uniformly colored features is referred to herein as = the number of uniformly colored particles. "Generally, the color images that will be analyzed, such as in this technical institute, are mostly uniform sentence colors. Many parts of the particle number. Packed in 9 color format or color space, such as: RGB format.. Separate color values for red, green, and blue, and service format, including separate values for hue, intensity, and saturation. , Means color image, is well-known in this technology. For the sake of simplicity, the application and the application park are explained throughout this case, unless you refer to Sun and Moon 'examples where RGB format will be used', please be aware that in most cases In other cases, other formats can also be used.

廳。影像之_單_彩色組份，可與咖影像之其他紐份予以早獨考慮；例如，綠组份可予以單獨考慮，在此辱技藝：熟知者。為處理及說明簡單起見，-僅包含綠組妨 I導得之影像，可^以考慮為—單色影像。在討論此-琴一組份導得之影像時，為說明簡單起見，一像素之值可稻作一灰階值，即使值在若干意義上為一，，綠階值，，。在灰階影像分析之領域，梯度之概念為熟知者，並且通常對應#像素值之函數之—次導數，㈣像素值在預定地點之速率變化。合理精於此項技藝者因此將會明白，在一單一彩色組份，諸如紅組份之均勻彩色之會包含具有零或近乎零梯度值之像素。非均二= -53-hall. The _single_color component of the image can be considered separately from the other components of the image; for example, the green component can be considered separately, in this shame technique: the familiar. For the sake of simplicity of processing and description,-only the images derived from the green group may be considered-monochrome images. When discussing the images derived from this component, for the sake of simplicity, the value of one pixel can be used as a gray level value, even if the value is one in several senses, the green level value, and so on. In the field of grayscale image analysis, the concept of gradient is well known, and usually corresponds to the second derivative of the function of #pixel value, the rate at which the pixel value changes at a predetermined location. Those skilled in the art will therefore understand that in a single color component, a uniform color such as the red component will contain pixels with zero or near zero gradient values. Non-uniformity = -53-

536919 ' A7 B7 五、發明説明（） 51 域，包括邊界及邊緣，通常將會包含具有顯著非零梯度值之像素。‘ 可使用於在灰階影像確定邊緣，包括在灰階影像確定至子像素準確度之較佳先前技藝方法，包括二次導數計算方法，其一般為使用一種高斯差（（difference of Gaussians，簡稱DOG)方法，計算像素強度函數之二次導數近似值。代表性先前技藝方法說明於下列美國專利，其揭示經予參考併入本案： 1. 授予Caspi之美國專利5,774,572號，說明一種自動目視檢查系統，其較佳為可操作卷積物體之平面數位灰階影像，而以一相關於高斯函數（Gaussian function)之二次導數之濾光函數，形成具有記號值之平面卷積影像。求出在相鄰相反記號值間之零交叉，藉以達成邊緣在物體之地點。 2. 授予Caspi等人之美國專利5,774,573號，說明一種目視檢查系統，其使用物體之平面數位灰階影像之卷積，而以一相關於高斯函數之二次導數之濾光函數，形成具有記號像素強度值之平面卷積影像。Caspi等人之卷積可利用一為正及一為負之高斯函數之差予以進行。彩色影像處理現請參照圖2，其為在圖1之CLRP單元170所包含之材料識別子單元170及材料輪廓子單元180，其一種較佳實施之簡化方塊圖示。 -54- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐)536919 'A7 B7 V. Description of the invention () 51 domains, including borders and edges, will usually contain pixels with significant non-zero gradient values. 'It can be used to determine edges in grayscale images, including the determination of subpixel accuracy in grayscale images. Prior art methods, including the second derivative calculation method, generally use a Gaussian difference ((Difference of Gaussians, for short) DOG) method to calculate the approximation of the second derivative of the pixel intensity function. Representative prior art methods are described in the following U.S. patents, the disclosures of which are incorporated herein by reference: 1. U.S. Patent No. 5,774,572 to Caspi, which describes an automatic visual inspection system , Which is preferably a planar digital grayscale image of a manipulable convolution object, and a planar convolution image with a sign value is formed by a filter function related to the second derivative of the Gaussian function. Zero crossings between adjacent opposite sign values to achieve the edge where the object is. 2. US Patent No. 5,774,573 to Caspi et al. Describes a visual inspection system that uses the convolution of a flat digital grayscale image of an object, and A filter function related to the second derivative of the Gaussian function is used to form a level with a marked pixel intensity value Convolutional image of the surface. The convolution of Caspi et al. Can be performed using the difference between a positive and a negative Gaussian function. For color image processing, please refer to FIG. 2, which is the material included in the CLRP unit 170 of FIG. 1. Identification sub-unit 170 and material contour sub-unit 180 are simplified block diagrams of a preferred implementation. -54- This paper size applies to China National Standard (CNS) A4 (210X297 mm)

裝訂Binding

52 536919 五、發明説明（通常，材料識別子單元17〇之任務4自輸人影像之彩色貧=產生儘量多關於輸入影像中之均勾彩色粒子數部位 Π纟材料識別子單元17G，如可能起因於存在半透明或部份透明塗層或覆蓋另一材料之表層，均勾彩色粒子數4，通常藉—特定材料或材料之組合予以識別。例如在丁以檢查〈物體為BGA _，代表性光學可辨別材料及材料，合另包銅導體，塗布或鍍敷金之銅導體，基 j ^及覆蓋一焊料掩模表層之任何上述材料之諸部份。請丁： ' T以仏查〈其他物體，可包括其他可辨別材料及材枓組合’諸如例如㈠艮。因此，材料識別子單元⑺ 之主要任務，為在輸入影像，並依據彩色資訊，識別均勾彩色粒子數之區域。如在整個本案說明書及申請專利範圍所使用，，半透明表層巧，包括例如半透明或部份透明焊料掩模，或例如半透明基片，通過其可見一導體位於一其相反侧。通常，均勻彩色粒子數之區域在本文稱作表示一特定材料〈區域’其通常為在_將行予以分析之輸人影際物體之影像時之情形。然而，請予察知，如以上所解釋，本發明不限於僅分析為實際物體之影像之輸入影像，並且在影像可能不是材料物體之影像之情形，使用"材科一巧係意在作為類比，並非意為限制。通常，材料輪廓單元18〇之任務為自輸入影像之彩色資 55 本紙張尺度咖A4規格(210X297公釐） 536919 A752 536919 V. Description of the invention (Usually, task 4 of the material identification subunit 170. 4 color deficiency of the input image = generate as much as possible about the average number of color particles in the input image. 纟纟 Material identification subunit 17G, if possible due to There is a semi-transparent or partially transparent coating or a surface layer covering another material, and the number of colored particles is 4, which is usually identified by a specific material or combination of materials. For example, in Ding Yi's inspection, the object is BGA _, representative optics Distinguishable materials and materials, together with copper conductors, copper conductors coated or plated with gold, substrates and parts of any of the above materials covering the surface of a solder mask. Please: "T to check <other objects , Can include other discernible materials and material combinations such as, for example, ㈠. Therefore, the main task of the material identification subunit 为 is to identify the area where the number of colored particles is checked based on the input image and based on the color information. As in the entire case As used in the specification and patent application, translucent surface layers include, for example, translucent or partially transparent solder masks, or, for example, translucent substrates It can be seen that a conductor is on the opposite side. Generally, a region of uniformly colored particles is referred to herein as representing a specific material "region", which is usually the case when an image of a human object is analyzed. However, please be aware that, as explained above, the present invention is not limited to input images that are only analyzed as images of actual objects, and in cases where the images may not be images of material objects, the use of "Materials Inc." is intended as an analogy It is not meant to be a limitation. Generally, the task of the material contour unit 18 is the color data from the input image. 55 paper size A4 size (210X297 mm) 536919 A7

訊及自材料識別子單元之區域間之邊界之指示以識別。 170 <輸出，產生均勻彩色粒子數，該等邊界較佳為在子像素位準予圖2之裝置較佳為包含下列： L 一材料分類單元171 ; 2· —彩色邊緣檢測器172 ; 3. —部位擴展單元173 ; 4· 一 cel構造單元174 ;以及 5· —輪廓化單元175。材料分類單元171較佳為接收彩色組份資料，一般為包含供輸入影像中每-像素之RGB值，並且較佳為^操= 將屬於一特定材料之大都為均勻彩色粒子數之區域分類。在一種較佳實施，材料分類單元171予以實施如一硬體 LUT，以便每一像素之分類直接依據該像素之彩色组份值，一般為RGB值，以便在與一特定材料或材料组合關聯之彩色空間内之RGB值將會與該材料關聯。較佳為，在以類似材料構成之許多物體之影像將行予以分析之代表性情形，使用在此項技藝所熟知之方法，分析代表性影像，可藉以建成LUT。請予察知，在此項技藝為熟知之分析代表性影像之若千適當方法，包括依據似直方圖像素彩色分布分析之簡單方法，並且可使用任何適當方法。供進行此種分析之一較隹方法，說明於下列參考 -56- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐）The boundary between the information and the material identification subunit is indicated. 170 < output, generating uniform number of colored particles, these boundaries are preferably at the sub-pixel level. The device of FIG. 2 preferably includes the following: L a material classification unit 171; 2 ·-color edge detector 172; 3. -Part expansion unit 173; 4.cel construction unit 174; and 5.-contouring unit 175. The material classification unit 171 preferably receives color component data, and generally includes RGB values for each pixel in the input image, and preferably ^ = classifies a region belonging to a specific material, which is mostly a uniform number of colored particles. In a preferred implementation, the material classification unit 171 is implemented as a hardware LUT, so that the classification of each pixel is directly based on the color component value of the pixel, generally an RGB value, in order to associate the color with a specific material or material combination. The RGB values in space will be associated with the material. Preferably, representative images of many objects composed of similar materials will be analyzed. Using methods well known in the art to analyze representative images, LUTs can be built. Please note that this technique is well known as a suitable method for analyzing representative images, including simple methods based on histogram pixel color distribution analysis, and any suitable method may be used. A comparative method for performing this analysis is described in the following reference. -56- This paper size applies to China National Standard (CNS) A4 (210X297 mm)

裝訂Binding

536919 A7 B7 五發明説明（ 54 貧料，其揭示經予參考併入本案：Dorin Comaniciu及Peter Meer，’’Distribution Free Decomposition of Multivariate Date'’， SPR’98 Invited Submission, Department of Electrical and Computer Engineering，Rutgers University，Piscataway，NJ 08855, USA。材料分類單元171較佳為輸出分類材料資訊，指示輸入影像中每一像素指定至一特定材料。彩色邊緣檢測器172較佳為接收與材料分類單元171相同輸入。彩色邊緣檢測器172較佳為可操作計算一近似一次導數，如在此項技藝所熟知，一般為藉應用一 Sobel運算符，一般為至HSI輸入影像之組份。Sobel運算符較佳為單獨應用至每一 HSI組份，然後並如以下所解釋，導得一合併結果。536919 A7 B7 Five Invention Notes (54 Poor material, whose disclosure is incorporated by reference into this case: Dorin Comaniciu and Peter Meer, `` Distribution Free Decomposition of Multivariate Date '', SPR'98 Invited Submission, Department of Electrical and Computer Engineering, Rutgers University, Piscataway, NJ 08855, USA. The material classification unit 171 preferably outputs classified material information indicating that each pixel in the input image is assigned to a specific material. The color edge detector 172 preferably receives the same material as the material classification unit 171. Input. The color edge detector 172 is preferably operable to calculate an approximate first derivative. As is well known in the art, it is generally to apply a Sobel operator, which is generally a component of the input image to the HSI. The Sobel operator is preferred For individual application to each HSI component, and then as explained below, a combined result is derived.

Sobel邊緣檢測在此項技藝為熟知者。函數0之梯度量級，係由0之偏導數在X及y方向之量級之和所示，其在Sobel估計在一點（i，j)予以約計如下： IΦ (i-1 J-0+2 φ (ij-l)+( φ (i-lj-l)+( φ (i-1 J+l)+2 φ (ij+l)+ φ (i+1 J+l))|+ I Φ (i-1 J-0+2 φ (i-l j)+( ψ (M J+l>( φ (i+1 j-l)+2 φ (i+1 J)+ φ (i+1 J+l))| 在本發明之Sobel邊緣檢測，較佳為使用一圓形3 X 3 邊緣檢測器，並經由影像卷積，根據以上公式計算2D梯度量級為垂直及水平梯度組份之絕對值之和。一般為，可使用下列核心供影像卷積： -57- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐） 536919 A7 B7 五、發明説明（） 55 1. 垂直梯度核心· 10 1 -2 0 2 -10 1 2. 水平梯度核心： 1 2 1 0 0 0 -1 -2 -1 彩色邊緣檢測器172然後較佳為計算一彩色邊緣為 Η，S，及I組份梯度之加權和如下： |GCOLOR|=(|GH|Relevance(I)WeightH+ |Gs|Relevance(I)WeightS+|Gi| WeghtI) 其中：Sobel edge detection is well known in the art. The magnitude of the gradient of function 0 is shown by the sum of the magnitudes of the partial derivatives of 0 in the X and y directions, which is estimated at Sobel's estimate at a point (i, j) as follows: IΦ (i-1 J-0 +2 φ (ij-l) + (φ (i-lj-l) + (φ (i-1 J + l) +2 φ (ij + l) + φ (i + 1 J + l)) | + I Φ (i-1 J-0 + 2 φ (il j) + (ψ (M J + l > (φ (i + 1 jl) +2 φ (i + 1 J) + φ (i + 1 J + l)) | In the Sobel edge detection of the present invention, it is preferred to use a circular 3 X 3 edge detector and image convolution to calculate the 2D gradient magnitude as the absolute value of the vertical and horizontal gradient components according to the above formula. In general, the following cores can be used for image convolution: -57- This paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm) 536919 A7 B7 V. Description of the invention () 55 1. Vertical gradient core · 10 1 -2 0 2 -10 1 2. Horizontal gradient core: 1 2 1 0 0 0 -1 -2 -1 The color edge detector 172 then preferably calculates a color edge as Η, S, and I components The weighted sum of the gradients is as follows: | GCOLOR | = (| GH | Relevance (I) WeightH + | Gs | Relevance (I) WeightS + | Gi | WeghtI) where:

Gcolor為彩色邊緣； GH為色相之梯度，Gcolor is the color edge; GH is the gradient of hue,

Gs為飽和之梯度；Gs is a saturated gradient;

Gi為強度之梯度；Gi is the gradient of intensity;

Relevance(l)為強度之彩色關聯性權重函數；Relevance (l) is the color correlation weight function of intensity;

WeightH為色相組份之權重；WeightH is the weight of the hue component;

Weights為飽和組份之權重；以及 Weightl為強度組份之權重。緊接上面所述諸因數之進一步討論，見於以上所提及 -58- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) 536919 A7 B7 五、發明説明（） 56 之 Philippe Pujas 及 Marie_Jose Aldon，"Robust Color Image Segmentation”，其揭示經予參考併入本案。現請參照圖3，其為操作彩色邊緣檢測器172之一種較佳方法之簡化流程圖示。參照以上討論，自可明白圖3 之流程圖。Weights are the weights of the saturated components; and Weightl is the weights of the intensity components. Following the further discussion of the factors mentioned above, see above mentioned -58- This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) 536919 A7 B7 V. Philippe's description of the invention () 56 Pujas and Marie_Jose Aldon, "Robust Color Image Segmentation", which is incorporated by reference in this case. Now refer to FIG. 3, which is a simplified flow diagram of a preferred method of operating the color edge detector 172. Refer to the discussion above You can understand the flowchart in Figure 3.

裝請回頭參照圖2，部位擴展單元173較佳為接收由材料分類單元171所產生之輸出，包含分類材料資訊，及彩色邊緣檢測器172所產生之輸出，包含彩色梯度資訊。部位擴展單元173較佳為可操作使包含如像素分類單元所識別均勻彩色粒子數之每一部位延伸儘量遠，以便在均勾彩色粒子數間之邊界儘量窄。 # 現請另外參照圖4，其為操作圖2之部位擴展單元173 之一種較佳方法之簡化流程圖示。在圖4之方法中，未指定之像素予以指定至具有低梯度之均勻彩色之相鄰部位，指示未指定像素迄至目前可能指定至具有未指定像素之最不清楚邊界之相鄰部位。圖4之方法較佳為包括下列步驟：使用如以上所說明之適當邊緣抽取運算符，在影像在每一像素計算梯度（步騾330 )。指示在圖4之方法之目前階段將行予以分類之未分類像素之梯度位準之目前梯度位準，予以設定至0 (步驟340)。較佳為進行下列步騾，供梯度少於或等於目前梯度位 -59- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) 536919 A7Please refer back to FIG. 2. The part expansion unit 173 preferably receives the output generated by the material classification unit 171, which includes classified material information, and the output generated by the color edge detector 172, which includes color gradient information. The part expansion unit 173 is preferably operable to extend each part including the number of uniformly colored particles as recognized by the pixel classification unit as far as possible so that the boundary between the number of uniformly colored particles is as narrow as possible. # Please also refer to FIG. 4, which is a simplified flowchart of a preferred method of operating the part expansion unit 173 of FIG. 2. In the method of FIG. 4, the unspecified pixels are assigned to the adjacent portions having a uniform gradient with a low gradient, indicating that the unspecified pixels have so far been assigned to the adjacent portion having the most unclear boundary of the unspecified pixels. The method of Figure 4 preferably includes the following steps: Calculate the gradient at each pixel in the image using the appropriate edge extraction operator as described above (step 330). The current gradient level indicating the gradient levels of the unclassified pixels to be classified at the current stage of the method of FIG. 4 is set to 0 (step 340). It is better to perform the following steps for the gradient to be less than or equal to the current gradient level -59- This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) 536919 A7

58 58536919 五、發明説明（供在灰p白影像作成此種子像素確定之適當裝置及方法說月於4又予Caspi之美國專利m4,5?2號，及授予 faSP1寺人之美國專利5,774,573號（在本文中也稱作 Caspi專利）’以上均已提及，並且其說明經予參考併入本案。請予察知，在本發明，可使用說明於Caspi專利之方法，進行在子像素位準，在單—組份灰階影像，諸如紅组份灰階影像，確定邊界之地點。如在此項技藝所熟知，特別是在以上所提及之c⑽扣專利，可使用適當高斯差（D〇G)運算符之卷積，計算二次導數，並因此確定邊界地點；特別是，可使用一大及一小運算符。較佳之核心包括3Χ1，5χ1，及5χ3核心。現請另外參照圖5 1 6 ’其分別示一較佳5χ5大核心及一較佳3 X3對應小核心之實例。圖5及6為自可明白。輪廓化單元175較佳為自cel構造單元174接收r， G，及B cel ’並自部位擴展單元173接收分段材料資訊，並且較佳為可操作自其產生彩色eel，表示在輸入影像在相鄰材料間之邊界。現請另外參照圖7，其為圖2之輪廓化單元175，其一種較佳實施之簡化方塊圖示。圖7之裝置較佳為包含下列： 1. 一材料cel建造器176，其較佳為接收分段材料資訊’並較佳為可操作自其產生以像素準確度表示邊界之材 61 - 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐） 536919 A758 58536919 V. Description of the invention (appropriate device and method for determining this seed pixel in gray and white images) U.S. Patent No. m4,5-2 issued to Caspi and U.S. Patent No. 5,774,573 issued to faSP1 temple (Also referred to herein as the Caspi patent) 'All of the above have been mentioned, and their descriptions are incorporated by reference into this case. Please be aware that in the present invention, the method described in the Caspi patent can be used to perform sub-pixel levels In single-component gray-scale images, such as red-component gray-scale images, determine the location of the boundary. As is well known in the art, especially in the c-patent patents mentioned above, an appropriate Gaussian difference (D 〇G) operator convolution, calculate the second derivative, and determine the boundary location accordingly; in particular, large and small operators can be used. The preferred cores include 3 × 1, 5 × 1, and 5 × 3 cores. Please refer to the figure separately 5 1 6 'shows examples of a preferred 5 × 5 large core and a preferred 3 X3 corresponding small core. Figures 5 and 6 are self-explanatory. The contouring unit 175 preferably receives r, G from the cel construction unit 174. , And B cel ' The exhibition unit 173 receives segmented material information, and is preferably operable to generate color eel from it, indicating the boundary between adjacent materials in the input image. Now please refer to FIG. 7, which is the contouring unit 175 of FIG. 2, A simplified block diagram of a preferred implementation thereof. The device of FIG. 7 preferably includes the following: 1. A material cel builder 176, which preferably receives segmented material information 'and is preferably operable to generate from it to Pixel accuracy indicates the material of the border 61-This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) 536919 A7

59 料 cel ; 材料cel建將行確定邊 2· —選擇器判定包絡建造器177，其繞每一造一包絡，諸如一 1像素包絡，指示在其内界之邵位；以及 3· —取佳輪廓選擇器178，較佳為自材料^丨建造器 176接收材料cel，及自選擇器判定包絡建造器Η? 包絡，以及R，G，及b cel資訊，並且較佳為可操作自其產生一最後彩色cel，表示在相鄰材料間之最佳邊界。請予察知，在最佳輪廓選擇器178確定邊界時，通常難以或不可能在邊緣之結點附近，以子像素準確度確定邊界’因此在結點附近較佳為使用輪廓化邊界。結點附近可以任何適當方式予以界定，諸如例如為在一結點之3像素正方形近鄰以内。在本發明之一種較佳實施例，根據在邊界兩侧之材料’判疋疋否使用子像素R, G，或B邊界’或在不靠近結點之區域是否使用像素㈣eel邊界。某些材料已知例如在紅色為最佳可辨別；如果此等材料在邊界二侧，則使用紅色子像素邊界使邊界固定在該地點。其他材料可能在 R，G，或B之任何者為不可辨別，在該情形，可例如以預定方式將RGB輸入加權，藉以產生一假彩色，以供邊界確定’或在該點將會使用一材料^丨邊界。現請參照圖8，其為操作圖7之最佳輪廓選擇器178， -62-59 material cel; material cel construction will determine the edge 2-the selector determines the envelope builder 177, which creates an envelope around each, such as a 1-pixel envelope, indicating the position in its inner boundary; and 3. · take The best profile selector 178, preferably from the material ^ 丨 the builder 176 receives the material cel, and the self-selector determines the envelope builder Η the envelope, and R, G, and b cel information, and is preferably operable from it. A final colored cel is generated, indicating the best boundary between adjacent materials. Please note that when the best contour selector 178 determines the boundary, it is usually difficult or impossible to determine the boundary with sub-pixel accuracy near the edges of the edge. Therefore, it is preferable to use a contoured boundary near the nodes. The vicinity of a node may be defined in any suitable manner, such as, for example, within the neighborhood of a 3-pixel square at a node. In a preferred embodiment of the present invention, it is determined whether the sub-pixel R, G, or B boundary is used or whether the pixel boundary is used in a region not close to the node according to the materials on both sides of the boundary. Some materials are known, for example, to be best discernible in red; if these materials are on either side of the boundary, a red subpixel boundary is used to fix the boundary at that location. Other materials may be indistinguishable at any of R, G, or B, in which case, the RGB input may be weighted in a predetermined manner, for example, to produce a false color for boundary determination 'or a material will be used at that point ^ 丨 border. Please refer to FIG. 8, which is the best contour selector 178 for operating FIG. 7, -62-

36919 A7 B736919 A7 B7

五、發明説明（參照以上討論，圖8 其一種較佳方法之簡化流程圖示之方法為自可明白。較佳為，如以上所說明之子像為許多輪廓元件或cel，每一 cei 之子像素準確度邊界分段；等於一像素尺寸之小向量。一側界定cel之地點及取向素準確度邊界予以表示指示在一單一像素附近每一 cel約略同等於長度大约界足—cel，包括在cel之每以及均勻彩色粒子數，其中如以上所說明，每-粒子數係與—種材料或材料組合相關，其一種較佳方法係如下：V. Description of the Invention (Refer to the discussion above. Figure 8 shows a simplified method of a preferred method. The method is self-explanatory. Preferably, as described above, the sub-images are a number of contour elements or cels, and the sub-pixels of each cei Segmentation of accuracy boundary; Small vector equal to one pixel size. The location of cel on one side and orientation element accuracy boundary are indicated to indicate that each cel near a single pixel is approximately equal to the length of approximately cel, including cel, The number of particles and the number of uniform colored particles, as described above, the number of particles per particle is related to a material or a combination of materials. A preferred method is as follows:

_說明一材料邊緣方向 0 ，供儲_存之代表性位編入一表，指示那些材料為在cel之每一侧在首先自Cel之第一端行進至cel之最後端時 ’如果負DOG之區域為至cei之左，則為否則為1 邊緣碼指示之方向，將像素之第一邊緣及最後邊緣編碼裝訂_Describe a material's edge direction of 0. Representative bits for storage are stored in a table indicating that those materials are on each side of cel when they first travel from the first end of Cel to the last end of cel. The area is to the left of cei, otherwise it is the direction indicated by the edge code, the first edge and the last edge of the pixel are coded and bound

第一邊緣在循環座標系統，與第一邊緣之cel之交點取後邊緣交點，在一循環座標系統，與最後邊緣之eel 之交點現請另外參照圖9 一 11，其為使用於暸解本發明之較佳 cel構造及術語表之簡化畫面圖示。圖9之圖示包含許多像素530及第二許多虛擬cei像素540。每一虛擬cei像 -63- 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐) 536919 A7 B7 五、發明説明（） 61 素540 —般為與每一像素530相同大小，並且每一虛擬 cel像素540中心在四像素530之一角周圍。在圖9中，僅意在作為實例，包含一中心虛擬cel像素545之相鄰像素之四像素530之每一像素，示為分別有既定之梯度值一 100，+20，+50，及+20。圖9之圖示也包含一 cel 550。cel 550之第一邊緣交點 560之地點，較佳為如圖示，在相鄰cel梯度值，例如+20 及一100之間，藉線性内插予以固定，俾如在此項技藝所熟知，估計零交叉點。同樣，cel 550之最後邊緣交點570 之地點，較佳為如圖示，在相鄰cel梯度值，例如+50及 -100之間，藉線性内插予以固定。圖10示一供說明cel 550之第一邊緣及最後邊緣，以及邊緣編號系統之較佳圓形座標系統，其中cel 550之頂，左，右，及底邊緣予以分別編號為0，1，2，及3。圖 11示一供說明邊緣碼，指示圖9及10之cel 550之方向之較佳方案。在四相鄰像素確定邊緣碼及方向基礎之DOG符號值之一種較佳方法，將會在一依據下列之表進行表查找： D〇G符號邊緣碼方向 ++ 7 0 ++ -+ 0 1 -64- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐) 536919 A7 B7 五、發明説明（ 62 ) ++ + - 1 0 ++ -- 3 0 ++ ++ 4 1 -1 -+ 4- 2 1 -r + - 6 1 卞 5 1 -+ ++ 5 0 + - 4- 6 0 \ -+- 4- 2 0 卞一 4- 4 0 Γ — ++ 3 1 -65- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐) 536919 A7 B7 五、發明説明（） -+ 1 1 +- 0 0 — 70 請予察知，在緊接上述之邊緣碼，碼6表示一鞍部，及碼7表示’'無cel ’·。根據圖10之編號系統，可自下列確定cel之第一及最後邊緣： -66 - 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐） 536919The first edge is in the circular coordinate system. The intersection point of the first edge with the cel is the rear edge intersection point. In a circular coordinate system, the intersection point with the eel of the last edge. Please refer to FIGS. 9-11. Simplified screen illustration of the better cel structure and glossary. The diagram of FIG. 9 includes a number of pixels 530 and a second number of virtual cei pixels 540. Each virtual cei image -63- This paper size applies Chinese National Standard (CNS) A4 specifications (210X 297 mm) 536919 A7 B7 V. Description of the invention () 61 Element 540-generally the same size as each pixel 530, and Each virtual cel pixel 540 is centered around one corner of the four pixels 530. In FIG. 9, only as an example, each pixel of four pixels 530 including adjacent pixels of a central virtual cel pixel 545 is shown as having predetermined gradient values of 100, +20, +50, and + 20. The illustration in Figure 9 also includes a cel 550. The location of the first edge intersection 560 of cel 550 is preferably as shown in the figure. The adjacent cel gradient values, such as +20 and -100, are fixed by linear interpolation. As is well known in the art, Estimate zero crossings. Similarly, the location of the last edge intersection 570 of cel 550 is preferably as shown in the figure, and the linear cel gradient values, such as between +50 and -100, are fixed by linear interpolation. Figure 10 shows a preferred circular coordinate system for explaining the first and last edges of the cel 550 and the edge numbering system. The top, left, right, and bottom edges of the cel 550 are numbered 0, 1, 2 respectively. , And 3. FIG. 11 shows a preferred scheme for explaining an edge code indicating the direction of cel 550 of FIGS. 9 and 10. A better method for determining the DOG symbol value of the edge code and direction based on four adjacent pixels is to perform a table lookup according to the following table: DOG symbol edge code direction ++ 7 0 ++-+ 0 1 -64- This paper size applies Chinese National Standard (CNS) A4 specification (210X297 mm) 536919 A7 B7 V. Description of the invention (62) ++ +-1 0 ++-3 0 ++ ++ 4 1 -1 -+ 4- 2 1 -r +-6 1 卞 5 1-+ ++ 5 0 +-4- 6 0 \-+-4- 2 0 卞一 4- 4 0 Γ — ++ 3 1 -65- This paper size applies Chinese National Standard (CNS) A4 specifications (210X297 mm) 536919 A7 B7 V. Description of the invention ()-+ 1 1 +-0 0 — 70 Please note that in the edge code immediately above, code 6 Indicates a saddle, and code 7 indicates `` no cel '. According to the numbering system of Fig. 10, the first and last edges of cel can be determined from the following: -66-This paper size applies the Chinese National Standard (CNS) A4 specification (210X 297 mm) 536919

邊緣碼 ο 1 ο 3 4 5 6 7 第一邊緣之編號 0 0 0 1 1 2 2 取後邊緣之編號 1 2 3 2 3 3 未界定未界定之在^發明之-種較佳實施例，如以上所說明予以計算之㈣邊:以：集二影像圖，表示在均勾彩色粒子數間邊界，並就母-eel指示均”色粒子數 =:二之任一侧。容易察知，如在上文所說明，使材於正…色粒子數相關’彩色cel影像圖包括關於存在 '行丁以檢查之物體之材料之識別及地點之資料。在本發明之-種較佳實施例，eel之收集較佳為包括在所檢查物體之特色之報告，其可依希望^另外處理外之報告較佳為包括—縮小之彩色圖，衫正行予以檢杏 •^物體心均句彩色粒子數之部位用以產生〜1，俾二:::: 完全說明’產生-彩色形態圖。另外，彩“d可予以分Edge code ο 1 ο 3 4 5 6 7 The number of the first edge 0 0 0 1 1 2 2 The number of the trailing edge 1 2 3 2 3 3 The edge of the calculation described above is as follows: Set two image maps, indicating the boundary between the number of uniformly-colored particles, and indicate the average-color number of mother-eel =: either side of two. As described above, the color cel image diagram related to the number of positive ... colored particles includes information about the identification and location of the material where the object exists for inspection. In a preferred embodiment of the present invention, eel The collection is preferably a report including the characteristics of the object being inspected, which can be processed as desired. ^ Other reports are preferably included-a reduced color map, which is checked by the shirt. The parts are used to generate ~ 1, 俾 2 :::: fully explain the 'generation-color morphological map. In addition, the color "d can be divided

-67- 65 536919 五、發明説明（析，以確定各不同塗層之形妝、土曰巧狀及地點，諸如部份透明表層’並在檢測為有缺陷時，輸出缺陷報主。現請參照圖12，其為圖1々少& <彩色表面缺陷子單元 19〇 ’其一種較佳實施之簡化 {规圖π。為參考簡便起見，彩色表面㈣子單元19G在本文有時稱作⑺·娜 190 ° 也稱作區域缺陷之表面缺陷，可予以說明為在隨機紋理二面内之局部異常，並且_般為另包括：殘留物，諸如在丁以檢查《物品I表面所保留之製程殘留物，在各不同二:及㈣’在表面物品之外形’包括特別在金屬 …不規律，擦痕’外來材料，汗點，以及各不同金屬 =屬塗層之氧化作用。表面缺陷較佳為予以分類為二類 -高對比缺陷’其通常包含具有明確界定之邊緣，在 /、上可測量強梯度之缺陷，諸如擦痕或凹點；以及 :低對比寬區域缺陷，其通常包含有不清楚邊緣，在其上可測量相對弱梯彦之缺阶了㈣度之缺fe，邊如水汙點或氧化部位。缺陷也可以其他方式予以分類，諸如：一彩色缺陷，其通常主要以彩部位為彩色缺陷之實例；収色不R於其周圍；氧化 -無彩色缺陷，或強度取向缺陷，亦即較亮或較暗於 Γ紙張尺歧财€家標準 -68- X 297公釐） 536919 A7 B7 五、發明説明（ 66 其周圍之缺陷，諸如例如凹點，擦痕，及汙點。通常，圖12之裝置尋求檢測如以上所說明之表面缺陷，即使在包含各種材料及紋理之外形複雜表面。較佳為，圖12之裝置予以設計為在一材料内操作，而不管其邊、彖並因此予以设计為接收一分段為材料及邊緣之輸入，諸如以上所說明材料輪廓子單元18〇所產生之輸出。車又佳為，圖12之裝置使用多解析度檢測，應用色及無彩色檢測器，在不同解析度，並行達成其結果。圖12之裝置可較佳予以構形為在不同靈敏度操作，供不同之材料，並在為距材料邊緣之距離之函數之不同靈敏度。較佳為，圖12之裝置產生所檢測表面缺陷之報告。較佳為，表面缺陷予以考慮為候選缺陷，並且報告伴有在予以檢查之影像之每一可能有缺陷地點之快照影像；報告及快照影像可如希望予以另外處理。因此，請予察知，在本發明，至圖12之裝置之較佳輸入，包括界定予以檢查之物體或影像之彩色影像資料，如材料識別子單it m所產生之材料資訊，及如材料輪廊單元180所產生之彩色cel資訊。較佳為，為供操作效率之目的，如在此項技藝所熟知圖12 i裝置較佳為部份或全部在特殊目的硬體予以實施，如果有其餘纟置，纟以軟體及通用目㈣體予以奋施。只 -69 --67- 65 536919 V. Description of the invention (analysis to determine the shape, makeup and location of various coatings, such as a partially transparent surface layer, and output a defect report to the owner when detected as defective. Referring to FIG. 12, it is a simplified & < colored surface defect sub-unit 19 of FIG. 1 which is a simplified implementation of a preferred implementation {plan π. For simplicity of reference, the colored surface sub-unit 19G is sometimes referred to herein. A surface defect called ⑺na 190 °, also known as a regional defect, can be explained as a local anomaly within the two sides of the random texture, and generally includes: residues, such as in Ding Yi Remaining process residues are different in two ways: and the shape of "outside surface objects" includes particularly in metals ... irregularities, scratches, foreign materials, sweat points, and the oxidation of different metals = coatings. Surface Defects are preferably classified as Class 2-High Contrast Defects, which typically include defects with clearly defined edges, measurable strong gradients on, such as scratches or pits; and: low contrast wide area defects, which Usually contains There are unclear edges, on which the relatively weak ladder can be measured, and the missing features such as water spots or oxidized parts can be measured. Defects can also be classified in other ways, such as: A color defect, which is usually mainly colored. Is an example of a color defect; the color is not R around it; oxidation-no color defect, or intensity orientation defect, that is, lighter or darker than Γ paper rule Qijia standard -68- X 297 mm) 536919 A7 B7 V. Description of the invention (66 Defects around it, such as, for example, pits, scratches, and stains. Generally, the device of FIG. 12 seeks to detect surface defects as described above, even if it is complex in shape other than including various materials and textures Preferably, the device of FIG. 12 is designed to operate within a material regardless of its edges, edges, and therefore is designed to receive an input segmented into materials and edges, such as the material profile subunit 18 described above 〇The output produced. The car is even better. The device in Figure 12 uses multi-resolution detection, and uses color and achromatic detectors to achieve its results in parallel at different resolutions. Figure 12 The device can preferably be configured to operate at different sensitivities for different materials and at different sensitivities as a function of distance from the edge of the material. Preferably, the device of Figure 12 produces a report of the detected surface defects. For this reason, surface defects are considered as candidate defects, and the report is accompanied by a snapshot image of each potentially defective location in the image being inspected; the report and the snapshot image may be processed separately if desired. Therefore, please be aware that in the present invention The preferred input to the device of FIG. 12 includes color image data defining the object or image to be inspected, such as material information generated by the material identification sub-item it m, and color cel information generated by the material gangway unit 180. Preferably, for the purpose of operating efficiency, as is well known in the art, the device of Figure 12i is preferably implemented partially or completely in special purpose hardware. If there are other settings, use software and general purpose. The body should work hard. Only -69-

67 五、發明説明（圖12之裝置較佳為包含下列單元：卜衫色1/彩色2造成單元彻，其較佳訊及材料資訊，並較佳為可操作自其導得在本文 1及彩色2 < 一導出彩色影像。夕在本發明較佳為，較佳彩色予以預先界定，每一材料之彩色缺陷，並且較佳彩色予以預先界定:= 測在每一材料之無彩彩色缺陷。彩色1影像然後予二足為供檢測無彩彩色缺陷之最佳影像，及彩色2同樣予以界定為供檢測彩色缺陷之最佳影像。每色: 影像及彩色2影像較佳為包含一大雜搶，其中每二像素根據其關聯材料予以指定一取自四可能灰階影像之—之像素.R ’ G ’ B ’及其HSI變換之一；所選擇之三聰變換〈-’車5佳為予以選擇為有效供預期存在於影像之材料。現請另外參照圖13，其為操作圖12之彩色w彩色2 造成單S 400，其一種較佳方法之簡化流程圖。參照以上討論，圖13之方法便自可明白。 2. —万向允許影像造成器4〇2，其接收彩色^丨資訊，並較佳為可操作產生二方向允許影像，〇kdir及子〇kdir，其較佳為用以控制圖12之裝置之另外操作。精於此項技藝者將會察知，在多材料影像應用敏感缺陷檢測器時，非常宜於避免源自材料邊緣之錯誤警告。 -70- ^紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公愛） 536919 A7 B7 五、發明説明（） 68 強梯度通常在材料邊緣予以測量；因此，在本發明，特別是關於表面缺陷檢測，宜於在或靠近材料邊緣禁止梯度測量。在本發明使用下列策略，以使可能求出靠近材料邊緣之缺陷，而不招致來自與材料邊緣關聯之梯度之錯誤警告。影像較佳為予以分為三主要類組：在邊緣，不作測量；靠近邊緣，僅在一平行於局部邊緣方向之方向作測量，此測量方向在本文稱作’’okdirn方向；在其他地方，進行正常測量。較佳為，使用一 okdir圖供全解析度測量，同時使用導得之’’子okdk"圖供子解析度測量。如本文中所稱”子解析度’’ 一詞，通常指使用子取樣影像，諸如一影像，其中僅使用在每一尺寸之每第η像素，諸如在每一尺寸之原始影像之每第5像素。請予察知，如在數學上所熟知，使用子取樣影像使影像”縮小 ”，由於距離縮小，而無像素值縮小，故增加某些計算值，諸如梯度值。以此方式，低對比缺陷可能變成更明顯。以下參照圖1 4更詳細說明造成okdir之方向允許影像造成器402之部份之一種較佳實施，以及以下參照圖 21 更詳細說明造成子okdir之部份者。 3.及4.多解析度梯度測量裝置，包含一子解析度通道低對比缺陷檢測器404，及一全解析度通道高對比缺陷檢測器406，各較佳為接收彩色1及彩色2影像，低對比檢 -71 - 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐)67 V. Description of the Invention (The device in FIG. 12 preferably includes the following units: The shirt color 1 / color 2 causes the unit to be complete, its best news and material information, and is preferably operable. Color 2 < A color image is derived. In the present invention, preferably, the preferred color is pre-defined, and the color defect of each material is pre-defined: = the colorless color defect of each material is measured. The color 1 image is then biped as the best image for detecting achromatic color defects, and the color 2 is also defined as the best image for detecting color defects. Each color: The image and color 2 images preferably include a large Miscellaneous grabbing, in which every two pixels are assigned a pixel taken from the four possible gray-scale images according to their associated materials. R 'G' B 'and one of its HSI transforms; the selected Sancon transform <-' car 5 Jiawei chose to be effective for the material expected to exist in the image. Now please refer to FIG. 13, which is a simplified flow chart of a better method for operating the single color S 400 of the color w color 2 of FIG. 12. Referring to the above discussion, Figure 13 method It is self-explanatory. 2. The universal allowable image generator 402 receives color information, and is preferably operable to generate two-direction allowable images, 〇kdir and sub-〇kdir, which are preferably used for Controls the additional operation of the device of Figure 12. Those skilled in the art will know that when applying sensitive defect detectors to multi-material images, it is very appropriate to avoid false warnings from the edges of the material. -70- ^ Paper size applies to China National Standard (CNS) A4 specification (210 X 297 public love) 536919 A7 B7 V. Description of the invention () 68 Strong gradients are usually measured at the edge of the material; therefore, in the present invention, especially regarding the detection of surface defects, it is appropriate to apply Gradient measurement is prohibited near the edge of the material. The following strategies are used in the present invention to make it possible to find defects near the edge of the material without incurring false warnings from gradients associated with the edge of the material. The images are preferably divided into three main groups : At the edge, no measurement is made; near the edge, the measurement is made only in a direction parallel to the local edge direction. This measurement direction is referred to herein as the `` okdirn direction ''; in other places The normal measurement is performed. Preferably, an okdir map is used for full resolution measurement, and the derived `` sub-okdk " map is used for sub-resolution measurement. As the term "sub-resolution" is referred to herein, Usually refers to the use of sub-sampled images, such as an image, where only every nth pixel at each size is used, such as every 5th pixel of the original image at each size. Please be aware that, as is known mathematically, use The sub-sampled image "shrinks" the image. As the distance is reduced, and no pixel value is reduced, some calculated values, such as gradient values, are added. In this way, low-contrast defects may become more apparent. Refer to Figures 14 and 4 for more details. The direction of causing the okdir allows a better implementation of the portion of the image generator 402, and the portion of the causing okdir is described in more detail below with reference to FIG. 3. and 4. A multi-resolution gradient measurement device, including a sub-resolution channel low-contrast defect detector 404 and a full-resolution channel high-contrast defect detector 406, each preferably receiving color 1 and color 2 images, Low Contrast Inspection -71-This paper size applies to China National Standard (CNS) A4 (210X297 mm)

裝訂 536919 A7 B7 五、發明説明（） 69 測檢測器404較佳為接收子okdir允許影像，及高對比檢測檢測器406較佳為接收okdir允許影像，每一低對比檢測檢測器404及高對比缺陷檢測器406為可操作進行梯度分析，並產生一缺陷報告。較佳為，採用總共四邊緣檢測器，一供在每一低對比缺陷檢測器404及高對比缺陷檢測器406之每一彩色1及彩色2影像。另外，較佳為在全解析度通道高對比缺陷檢測器406僅採用二小缺陷檢測器，一供每一彩色1及彩色2影像。 5. —通道報告合併器408，其較佳為可操作自每一低對比缺陷檢測器404及高對比缺陷檢測器406接收報告，並自其產生一適當合併報告。現請參照圖14，其為圖12之方向允許影像造成器402 之第一部份，其一種較佳實施之簡化方塊圖示。圖14中所例示方向允許影像造成器402之部份為造成okdir之部份；以下參照圖21說明造成子okdir之部份。圖14之裝置較佳為包含邊緣允許影像造成器410，其較佳為接收彩色cel資訊作為輸入，並較佳為可操作自其導得邊緣允許影像。在邊緣允許影像，’’碰觸”任何cel之所有像素，較佳予以標記為邊緣像素，而所有其他像素任令不予以標記，表示材料像素；因此造成邊緣允許影像如二進制影像。現請另外參照圖15，其為一影像像素及相鄰虛擬cel -72- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐）Binding 536919 A7 B7 V. Description of the Invention (69) The detection detector 404 is preferably to receive the okdir allowed image, and the high contrast detection detector 406 is preferably to receive the okdir allowed image. The defect detector 406 is operable to perform gradient analysis and generate a defect report. Preferably, a total of four edge detectors are used, one for each of the color 1 and color 2 images of each of the low contrast defect detector 404 and the high contrast defect detector 406. In addition, it is preferable that in the full-resolution channel high-contrast defect detector 406, only two small defect detectors are used, one for each color 1 and color 2 images. 5. A channel report combiner 408, which is preferably operable to receive reports from each of the low-contrast defect detector 404 and the high-contrast defect detector 406 and generate an appropriate consolidated report therefrom. Please refer to FIG. 14, which is a simplified block diagram of a preferred implementation of the first part of the image generator 402 in the direction of FIG. 12. The portion of the direction allowing image generator 402 illustrated in FIG. 14 is the portion causing the okdir; the portion causing the sub-okdir will be described below with reference to FIG. 21. The device of FIG. 14 preferably includes an edge-allowed image generator 410, which preferably receives color cel information as an input, and is preferably operable to derive an edge-allowed image from it. Allowing images at the edges, "touching" all the pixels of any cel, preferably marked as edge pixels, while all other pixels are left unlabeled, indicating material pixels; therefore, edges allow images such as binary images. Please also add another Referring to FIG. 15, it is an image pixel and adjacent virtual cel-72- This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm)

裝訂 536919 A7 B7 五、發明説明（像素之簡化畫面圖示，用於瞭解圖14之裝置之操作。一般為，考慮中心在繞影像像素690周圍之四虛擬cel像素，諸如虛擬cel像素700，710，720，及730之近鄰，供每一影像像素，諸如影像像素690。虛擬cel像素 700，710，720，及730 —般為根據其位置分別予以編號為 0，1，2，及 3。Binding 536919 A7 B7 V. Description of the invention (Simplified picture illustration of pixels, used to understand the operation of the device of Figure 14. Generally, consider four virtual cel pixels centered around the image pixel 690, such as virtual cel pixels 700, 710 , 720, and 730 are close neighbors for each image pixel, such as image pixel 690. Virtual cel pixels 700, 710, 720, and 730 are generally numbered 0, 1, 2, and 3 according to their positions, respectively.

裝如果無cel見於四相鄰虛擬cel像素700，710，720，及730之任何者，影像像素690不考慮為邊緣像素。如果一 cel見於四相鄰虛擬cel像素700，710，720，及730 之任何者，影像像素690予以考慮為邊緣像素，除非如圖 15中所指示，所有此等cel為除外之cel 740。根據圖10 及其以上說明，除外之cel 740包含：訂 # -73-本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐) 536919 A7 B7 五、發明説明（虛擬cel圖號參考圖號 700 710 720 730 請回頭參照圖14 除外之cel圖號 0 1 4 5 圖 14之裝置較佳為包含一形態控制單元412，一般為予以實施為一 LUT，其較佳為可操作提供控制資訊至許多腐敍單元414，一般為包含3腐蚀單元 414。如在此項技藝所熟知，每一 3腐蝕單元414較佳為可操作應用腐蝕至邊緣允許影像造成器41〇所造成之邊緣允弄影像。一般為，第一腐蝕單元414應用2位準之腐蝕，並產生一允許2圖；第二腐蝕單元414應用*更多位準之腐蝕，並產生一允許6圖；以及第三腐蝕單元々Μ 應用4更多位準之腐蝕，並產生一允許1〇圖，一般為，除了供其所接收之輸入及控制資訊，三腐蝕單元414各為完全相同。 … 較佳為，根據一有一值之腐蝕鍵實施每一腐蝕步驟. 腐蝕鍵值之一種可能實例如下："occcc丁 cc〇〇”。較佳為，腐蝕鍵之每一字符表示一單—腐蝕步驟，自左至=祛取，每一步驟較佳為根據一腐蝕運算符予以解釋如下/貝 __ -74· I紙張尺度適用中國國巧準(CNs)人4規格⑽/士公复)· --------^ 536919 A7 B7If no cel is seen in any of the four adjacent virtual cel pixels 700, 710, 720, and 730, the image pixel 690 is not considered as an edge pixel. If a cel is found in any of the four adjacent virtual cel pixels 700, 710, 720, and 730, the image pixel 690 is considered as an edge pixel, except as indicated in FIG. 15, all such cels are cel 740 except. According to Figure 10 and above, the excluded cel 740 includes: Order # -73- This paper size applies Chinese National Standard (CNS) A4 specifications (210X 297 mm) 536919 A7 B7 V. Description of the invention (virtual cel drawing number reference Drawing number 700 710 720 730 Please refer back to Figure 14 except for cel drawing number 0 1 4 5 The device of Figure 14 preferably includes a form control unit 412, which is generally implemented as a LUT, which is preferably operable to provide control Information to many corruption units 414, generally including 3 corrosion units 414. As is well known in the art, each 3 corrosion unit 414 is preferably operable to apply corrosion to the edge to allow the edge caused by the image generator 41. Generally, the first etching unit 414 applies 2 levels of corrosion and generates a permit 2 map; the second etching unit 414 applies * more levels of corrosion and generates a permit 6 map; and the third corrosion The unit MM applies 4 more levels of corrosion and generates an allowable 10 map. Generally, except for the input and control information it receives, the three corrosion units 414 are all the same.… Preferably, according to a Have a value The corrosion key performs each corrosion step. One possible example of the corrosion key value is as follows: " occcc 丁 cc〇〇 ". Preferably, each character of the corrosion key represents a single-corrosion step, from left to = remove Each step is preferably explained according to an erosion operator as follows: / Paper __ -74 · I paper size applies to China National Standards (CNs) person 4 specifications ⑽ / Shi Gongfu) ------- -^ 536919 A7 B7

五、發明説明（ 72 鍵字符腐钱運算符之說明 0 不腐蝕，輸出=輸入 S 保存，不耗用開口端 C 不間斷連接性 N 允許連接性間斷 T 吃掉小凸起部，藉以使邊緣平滑較佳為，所有腐蝕運算符包含正交、中心作喜、堤异付，邓即如果素〈四一階近鄰之-巳被吃掉，耗用一中心傻去、運异符。請予察知，允許2圖較佳為藉操作”〇c„ 圖另藉操作”CCCT”，及允許1G圖另藉操作㈣。，產I。6 精於此項技藝者將會察知，較佳為包括,,G"腐#步驟，以供客易某些硬體實施。允許圖較佳為用以在τ自其確定邊緣方向之邊緣内延伸範圍。腐蝕操作會增加原始邊緣允許影像之大小，因此允許在自邊緣之較大距離確定邊緣檢測；較佳為，可造成夕至1〇像素寬度之okdir包絡。 ° 圖14之裝置也較佳為包括一 okdir圖計算單元416，其可操作產生okdir影像。現請另外參照圖16，其為操作 okchr圖計算單元416之一種較佳方法之簡化流程圖示。圖16之方法較佳為包含下列步驟·· 自3允許影像當中（步驟418)確定一關聯允許影像。現請另外參照圖17，其為圖16之步驟418，其一種較V. Description of the invention (Description of the 72-key character rotten money operator 0 Non-corrosive, Output = Input S Save, Do not consume open end C Uninterrupted connectivity N Allow connectivity discontinuity T Eat up small protrusions to make edges Smoothing is better. All corrosion operators include orthogonality, center-happiness, and bank-difference payment. If Deng is eaten if the prime <fourth-order nearest neighbor- 巳 is consumed, it takes one center to go silly and transport the different characters. Please check It is known that allowing 2 pictures to borrow operations "〇c" Figure borrows operations "CCCT" and 1G pictures borrowing operations ㈣., I. 6 Those skilled in this art will know that it is better to include , G " 腐 # steps for easy implementation by some hardware. The allowable map is preferably used to extend within the edge where τ determines the edge direction. Corrosion operation will increase the size of the original edge allowed image, so Allows the edge detection to be determined at a large distance from the edge; preferably, it can cause an okdir envelope with a width of 10 to 10 pixels. ° The device of FIG. 14 also preferably includes an okdir map calculation unit 416, which is operable to generate okdir Image, please refer to Figure 16 separately, It is a simplified flow diagram of a preferred method of operating the okchr graph calculation unit 416. The method of FIG. 16 preferably includes the following steps ... Determine an associated permitted image from 3 permitted images (step 418). Please refer to it separately FIG. 17 shows step 418 of FIG.

裝訂 -75-Binding -75-

536919 五、發明説明（ A7 B7536919 V. Description of the invention (A7 B7

佳實施 < 簡化流程圖。下列定義可使用於瞭解圖17之方法：一包絡像素·存在於一 5 X 5矩陣之邊界之像素；一 okdir位址·包絡及圖碼，如以下所說明；一包絡：一所選擇允許影像包絡像素之位元圖，所選擇允許影像係自允許2影像，允許6影像，及允許ι〇影像當中予以選擇，較佳為根據圖17之方法；一圖碼，一代碼，一般為實施為_ 2位元碼，說明一所選擇允許圖，一般為根據下列解釋： ilil解釋 00 中心像素為邊緣像素 01 包絡取自允許2影像 02 包絡取自允許6影像 03 包絡取自允許10影像圖Π之方法較佳為接收包含允許2影像，允許6影像，及允許10影像之3允許影像作為輸入，並較佳為包含下列步驟：檢查允許2影像 < 中心像素是否為邊緣像素（步騾 420)。如為是，okdir位址予以設定至〇 (步驟422);亦即，包絡及圖碼均予以設定至〇，並且圖n之方法終止。否則’檢查允許2影像之包絡像素是否包括至少邊緣 -76- 536919Best Practice < Simplified Flowchart. The following definitions can be used to understand the method of FIG. 17: An envelope pixel. A pixel that exists at the boundary of a 5 X 5 matrix. An okdir address. Envelope and figure code, as described below. An envelope: a selected allowed image Envelope pixel bitmap, the selected allowed image is selected from allowed 2 images, allowed 6 images, and allowed 0 images, preferably according to the method of Figure 17; a picture code, a code, generally implemented as _ 2 digit code, indicating a selected allowed map, generally based on the following interpretation: ilil explained 00 center pixel is edge pixel 01 envelope is taken from allowed 2 images 02 envelope is taken from allowed 6 images 03 envelope is taken from allowed 10 images The method preferably receives as input including Allow 2 images, Allow 6 images, and Allow 10 images 3, and preferably include the following steps: Check if Allow 2 images < center pixel is edge pixel (step 420 ). If yes, the okdir address is set to 0 (step 422); that is, both the envelope and the figure are set to 0, and the method of figure n is terminated. Otherwise ’, check if the envelope pixels allowed for 2 images include at least edges -76- 536919

74 像素（步驟424)。如為是，包絡予以設定相等於允許2包絡，圖碼予以設定等於01 (步驟426)，並且圖17之方法終止。口則，檢查允許6影像之包絡像素是否包括至少邊緣像素（步騾428)。如為是，包絡予以設定相等於允許6包絡，圖碼予以設定等於02 (步騾43〇)，並且圖17之方法終止。否則，包絡予以設定等於允許1〇包絡，並且圖碼予以設定等於03 (步驟432)。通常，精於此項技藝者將會察知，圖17之方法可操作選擇最小允許圖之包絡，其包括邊緣像素，除非允許2 影像之中心像素為邊緣像素，在該情形，不選擇包絡。請回頭參照圖16，計算〇kdir (步驟419)。為供如以上所說明效率炙原因，較佳為以硬體在LUT實施步騾 W9。現請另外參照圖ls，其為圖16之步驟4丨9，其— 種較佳實施之簡化流程圖示。目18之方法較佳為包括下列步驟：檢查okdir位址是否等於〇(步驟4M);如為是，处出也予以設定等於0 (步驟436 )。檢查疋否佔用所有包絡像素（步驟43 8 );如為是予以設足等於15，一般予以表示為十六進制數〇xF 440)。 ’ 〇kdir (步驟74 pixels (step 424). If so, the envelope is set equal to 2 envelopes allowed, the pattern code is set equal to 01 (step 426), and the method of FIG. 17 is terminated. In general, check whether the envelope pixels of the allowed 6 images include at least edge pixels (step 428). If yes, the envelope is set equal to 6 envelopes allowed, the figure code is set equal to 02 (step 43), and the method of FIG. 17 is terminated. Otherwise, the envelope is set equal to allowing 10 envelopes, and the pattern code is set equal to 03 (step 432). In general, those skilled in this art will know that the method of FIG. 17 is operable to select the envelope of the smallest allowed map, which includes edge pixels, unless the center pixel of 2 images is allowed to be edge pixels. In this case, the envelope is not selected. Please refer back to FIG. 16 to calculate 0kdir (step 419). For reasons of efficiency as described above, it is preferable to implement step W9 in the LUT by hardware. Now please refer to FIG. Ls, which is step 4 丨 9 of FIG. 16, which is a simplified flow diagram of a preferred implementation. The method of item 18 preferably includes the following steps: Check if the okdir address is equal to 0 (step 4M); if it is, the processing is also set equal to 0 (step 436). Check whether all envelope pixels are occupied (step 43 8); if it is set to be equal to 15, it is generally expressed as a hexadecimal number 0xF 440). ’〇kdir (step

裝 -77-Equipment -77-

536919 A7 B7 五、發明説明（） 75 允許影像然後予以腐蝕至骨架頭（步騾442 );骨架頭為一在腐蝕後保持之單一隔離像素。現請另外參照圖19，其為導致骨架頭452之腐蝕過程，其一種實例之簡化畫面圖示。請回頭參照圖18，依據關於骨架頭之數之三可能情形之一採取動作（步騾444): 如果有1骨架頭，okdir為垂直於一連接骨架頭及展開之中心像素之向量（步驟446)。okdir方向之此情形在圖19 中示如參考圖號454。如果有2骨架頭，okdir為一在二骨架頭之間所繪之線之方向（步騾448 )。如果有3或更多骨架頭，okdir予以設定至0，指示將不取測量。現請另外參照圖20，其為一示okdir值，對應測量方向，及一供測量之較佳梯度運算符之表。在圖20之表中，如在該項技藝所熟知，”LinearDiff"指一線性微差運算符，而’’HalfSobel，’指一修改Sobel運算符，名為半Sobel運算符，其在以下予以更詳細說明。圖20之表在其他方面為自可明白。現請參照圖21，其為操作圖12之方向允許影像造成器 402之第二部份，其一種較佳方法之簡化流程圖示；圖 21之方法係針對產生子okdir影像。 -78- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐）536919 A7 B7 V. Description of the invention () 75 Allow the image to be etched to the skeleton head (step 442); the skeleton head is a single isolated pixel maintained after the erosion. Please refer additionally to FIG. 19, which is a simplified screen diagram of an example of the corrosion process leading to the skeleton head 452. Please refer back to FIG. 18, and take an action according to one of the three possible situations regarding the number of skeleton heads (step 444): if there is 1 skeleton head, okdir is a vector perpendicular to a connected skeleton head and the unfolded central pixel (step 446 ). This situation in the okdir direction is illustrated in FIG. 19 as referenced to reference number 454. If there are 2 skeleton heads, okdir is the direction of a line drawn between the two skeleton heads (step 448). If there are 3 or more skeletal heads, okdir is set to 0, indicating that no measurement will be taken. Please refer to FIG. 20 additionally, which is a table showing okdir values, corresponding measurement directions, and a preferred gradient operator for measurement. In the table of FIG. 20, as is well known in the art, "LinearDiff" refers to a linear difference operator, and "HalfSobel," refers to a modified Sobel operator, called a semi-Sobel operator, which is given below In more detail, the table in FIG. 20 is self-explanatory in other respects. Please refer to FIG. 21, which is a simplified flowchart of a preferred method for operating the second part of the image generator 402 in the direction of FIG. 12 ; The method in Figure 21 is for generating sub-okdir images. -78- This paper size applies to China National Standard (CNS) A4 (210X297 mm)

裝訂Binding

A7 B7A7 B7

536919 五、發明説明（ 76 圖21之方法較佳為包含下列步騾：接收一如以上參照圖14所說明所產生之允許1〇影像作為輸入，並予以擴展（步騾456)。較佳為，擴展允許 1〇影像包括進行5X5展開，並且如果在擴展之任一像素為邊緣像素1定展開之中^像素為邊緣像素。如此所產生心影像在本文稱作”基本允許”影像。以k較佳為等於u或η，進行基本允許影像之匕χ让展開（步驟458)。然後自< kxk展開之9預定像素，造成虛板3 X 3展開（步驟46〇 )。現請另外參照圖，其例示仪11X 11展開之預定像素之一種較佳選擇，請予察知，可作成預足像素之類似選擇，供21X21展開，或供另一展開。另请予察知，自kxk之9像素之選擇，包含子取樣之一種形式。患用組喊探’俾自虛擬3 X 3展開確定子〇kdir值（步 T 462)。現請另外參照圖23A，其為一表，例示供使用万、圖21之步驟462之較佳試探，並請參照圖MR，其為簡化畫面圖不’示供3x3虛擬展開之9像素之命名協足圖23B之名稱使用於圖23A。圖23A及23B為自可明白。現凊參照圖24，其為操作圖12之裝置之一部份，其一種較佳方法之簡化流程圖示。圖24之流程圖包含操作圖 12 <全解析度通道4〇6，子解析度通道，及通道報告536919 V. Description of the invention (76 The method of FIG. 21 preferably includes the following steps: Receive as input the allowable 10 image as described above with reference to FIG. 14 and expand it (step 456). It is preferably The expansion allows 10 images to include 5X5 expansion, and if any of the pixels in the expansion are edge pixels, ^ pixels are edge pixels. The heart image generated in this way is referred to herein as "basically allowed" images. It is preferably equal to u or η, and the expansion of the image is basically allowed (step 458). Then, from 9 predetermined pixels of < kxk expansion, the virtual plate is expanded by 3 X 3 (step 46). Please refer to it separately Figure, which exemplifies a preferred selection of the predetermined pixels of the 11X 11 expansion, please know that a similar selection of pre-footed pixels can be made for 21X21 expansion, or another expansion. Please also know that from 9 pixels of kxk The selection includes a form of sub-sampling. The patient group calls out '俾 from the virtual 3 X 3 expansion to determine the sub-kdir value (step T 462). Now please refer to FIG. 23A, which is a table illustrating examples for use. Step 46 of Figure 21 2 is a better test, and please refer to FIG. MR, which is not shown to simplify the screen picture. The name of 9 pixels for 3x3 virtual expansion is shown in FIG. 23B. Figures 23A and 23B are self-explanatory.凊 Referring to FIG. 24, which is a part of the device for operating FIG. 12, and a simplified flow chart of a preferred method thereof. The flowchart of FIG. 24 includes operation of FIG. 12 < Full resolution channel 406, sub-resolution Channels, and channel reports

裝訂 -79- 536919 A7 B7 五、發明説明（合併器408之一種較佳方法。關於圖12之上述部份之圖 24之流程圖之諸部份，在圖 24中由參考圖號404， 406，及408予以概括指示。較佳為，如以上參照圖12所說明，全解柝度通道406 及子解析度通道404均接收彩色1影像及彩色2影像作為輸入。圖24之方法較佳為包含下列步騾： 1.關於全解析度通道406之步騾：較佳為以一 3 X 3鏡高斯過濾使輸入影像平滑（步騾 464 )。如在此項技藝所熟知，高斯過濾可用以使影像平滑，並減低其中之紋理效應。請予察知，高斯過濾有一缺點，因為鄰近邊緣之像素可能受相鄰邊緣像素所影響，有效使邊緣向材料擴展；此問題可稱作”邊緣效應’’。在本發明，將高斯運算符與鏡運算符合併，使在3 X 3展開之像素變換為在3 X 3展開之相反側之像素之鏡高斯運算符，用以克服在先前技藝所已知之邊緣效應，並從而以一種不受關於邊界之影像資料所影響之方式，使關於非邊界部位之影像資料平滑。現請另外參照圖25，其為圖24之步驟464，其一種較佳實施之簡化流程圖，其中Gauss (i，j)表示應用一標準高斯函數。圖25之方法為自可明白。請回至圖24，應用一適當梯度運算符（步驟466 )，梯 -80- 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐)Binding-79- 536919 A7 B7 V. Description of the Invention (A preferred method of the combiner 408. Regarding the parts of the flowchart of FIG. 24 of the above part of FIG. 12, reference is made to the drawing numbers 404, 406 , And 408 are general instructions. Preferably, as explained above with reference to FIG. 12, the full resolution channel 406 and the sub-resolution channel 404 both receive color 1 image and color 2 image as input. The method of FIG. 24 is preferably Contains the following steps: 1. Steps on the full resolution channel 406: It is preferred to smooth the input image with a 3 X 3 mirror Gaussian filter (step 464). As is well known in the art, Gaussian filtering can be used to Smooth the image and reduce the texture effect in it. Please be aware that Gaussian filtering has a disadvantage, because the pixels of adjacent edges may be affected by the pixels of adjacent edges, which effectively expands the edge to the material; this problem can be called "edge effect" '. In the present invention, the Gaussian operator is matched with the mirror operation, and the pixel in the 3 X 3 expansion is transformed into the mirror Gaussian operator in the pixel on the opposite side of the 3 X 3 expansion to overcome the previous art. know Edge effect, and thus smooth the image data about non-boundary parts in a way that is not affected by the image data about the boundary. Please refer to FIG. 25 additionally, which is step 464 of FIG. 24, which is a preferred implementation Simplified flowchart, where Gauss (i, j) represents the application of a standard Gaussian function. The method in Figure 25 is self-explanatory. Please return to Figure 24 and apply an appropriate gradient operator (step 466). Paper size applies to China National Standard (CNS) A4 (210X 297 mm)

裝訂Binding

536919 A7 B7 五、發明説明（度運算符一般為包含一在材料部位内之半Sobd檢測器，及一沿邊緣之線性微差檢測器（LinearDiff)。在本發明，應用線性微差檢測器，較佳為包含在okdk 方向或在子okdir方向之測量，在子解析度通道之情形，在短向量之頭"b”與短向量之尾”a”間之像素值之差。線性微差檢測器之值，較佳為在’’b"之值減在之值。如在此項技藝所熟知，並如在本文所說明，Sobel運算符為一使一沿主水平及垂直軸線之灰度級導數合併，而以沿對角之灰度級推導卷積之梯度運算符。如以下所說明，在圖24之方法，Sobel運算符使用在子解析度通道。Sobel 運算符一般為可予以實施作為二單獨應用運算符，一水平運算符諸如： 10-1 2 0-2 1 0 -1 及一垂直運算符諸如： 1 2 1 0 0 0 -1 -2 -1 在本發明，一稱作半Sobel運算符之改進Sobel型運算符，較佳為使用於步驟466，目的為增加Sobel運算符檢測局部最小值及或最大值之特性。一般為，半Sobel水平 -81 - 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐)536919 A7 B7 V. Description of the invention (The degree operator generally includes a half Sobd detector in the material part, and a linear difference detector (LinearDiff) along the edge. In the present invention, a linear difference detector is applied. It is preferable to include the measurement in the okdk direction or the sub-okdir direction, in the case of the sub-resolution channel, the difference in pixel values between the head of the short vector " b "and the tail of the short vector" a ". Linear slight difference The value of the detector is preferably the value minus the value of "b ". As is well known in the art, and as explained herein, the Sobel operator is a gray scale along the main horizontal and vertical axis. Degree-level derivatives are combined, and the gradient operator of the convolution is derived at the gray level along the diagonal. As explained below, in the method of Figure 24, the Sobel operator is used in the sub-resolution channel. The Sobel operator is generally Implemented as two separate application operators, a horizontal operator such as: 10-1 2 0-2 1 0 -1 and a vertical operator such as: 1 2 1 0 0 0 -1 -2 -1 In the present invention, one is called Improved Sobel operator as a semi-Sobel operator, preferably In step 466, for the purpose of increasing the detection characteristics of Sobel operator and a local minimum or maximum of typically half Sobel horizontal -81--. This paper is suitable China National Standard Scale (CNS) A4 size (210X 297 mm)

裝訂 t 536919 A7Binding t 536919 A7

536919 A7 B7 五、發明説明（） 80536919 A7 B7 V. Description of the invention () 80

Grad=DifO+Difl 如果DifO及Difl之符號不同，可有一局部極端值，在或靠近C像素包含一局部最小值或一局部最大值。步騾466之輸出包含水平及垂直梯度影像，單獨自彩色 1及彩色2影像所獲得。然後使梯度正常化（步騾468 )，較佳為依據距邊緣之距離，以一正常化運算符，並較佳為單獨供DifO及Difl。在若干材料同時操作表面檢測器時，因為每一材料有一不同之固有紋理而出現問題，造成一不同位準之梯度，有時稱作”雜訊’’。指示在一材料之缺陷之梯度雜訊位準，在一第二材料可能視為正常紋理。請予察知，可採用各種不同方法克服在不同材料之不同紋理之問題。一種方法將為應用自適應界限。可代之為在一其時予以應用至如此所獲得之正常化梯度之界限内，提供一單獨之正常化函數供每一材料。在本發明之步驟468，較佳為使用梯度正常化。供每一材料，預先設定一適當正常化函數，正常化函數予以拾取以獲得均勻結果。較佳為適當正常化函數可予以確定如下：獲得所有材料之梯度之直方圖；確定一中間梯度值，供每一材料；以及如在此項技藝所熟知，應用正常化操作，以獲得正常化函數。在步驟468，然後較佳為計算正常化梯度如下： -83- 本紙張尺度適用中國國家標準(CNS) A4規格(210X 297公釐) 536919 A7 B7 五、發明説明（） 81Grad = DifO + Difl If the signs of DifO and Difl are different, there may be a local extreme value, which includes a local minimum value or a local maximum value at or near the C pixel. The output of step 466 includes horizontal and vertical gradient images, which are obtained separately from color 1 and color 2 images. The gradient is then normalized (step 468), preferably based on the distance from the edge, with a normalization operator, and preferably for DifO and Difl separately. When several materials operate the surface detector at the same time, problems occur because each material has a different inherent texture, resulting in a gradient of different levels, sometimes referred to as "noise." A gradient gradient indicating a defect in a material. The signal level may be regarded as a normal texture in a second material. Please be aware that various methods can be used to overcome the problem of different textures in different materials. One method will be to apply adaptive boundaries. It is applied to the limit of the normalization gradient thus obtained, and a separate normalization function is provided for each material. In step 468 of the present invention, gradient normalization is preferably used. For each material, a Appropriate normalization function, which is picked to obtain uniform results. Preferably, the appropriate normalization function can be determined as follows: obtain the histogram of the gradient of all materials; determine an intermediate gradient value for each material; and as in As is well known in the art, a normalization operation is applied to obtain a normalization function. In step 468, it is then preferable to calculate a normalization ladder. As follows: -83- This paper scales applicable Chinese National Standard (CNS) A4 size (210X 297 mm) 536919 A7 B7 V. invention is described in () 81

NormalizedGrad=NormFunction(material)* Gradient* EdgeFactor(NarrowArea) 其中：NormalizedGrad = NormFunction (material) * Gradient * EdgeFactor (NarrowArea) where:

NormalizedGrad為正常化梯度值；NormalizedGrad is the normalized gradient value;

NormFunction(material)為材料之正常化函數，其可如以上所說明予以確定；NormFunction (material) is the normalization function of the material, which can be determined as described above;

Gradient為輸入梯度值；及EdgeFactor(NarrowArea)為一因數，通常範圍在0與1 之間，並選擇為給予較少權重至靠近邊緣之梯度； EdgeFactor(NarrowArea)較佳為靠近邊緣例如具有 0.8之值。其計算如以上所說明之梯度，予以分析，分類，及界限ed (步驟470)。現請另外參照圖26，其為圖25之步騾 470之一部份，其一種較佳實施之簡化流程圖示，包括分析及分類。在圖26中，DifO及Dif 1較佳為如以上參照步驟466所說明。請予察知，如以上所說明，在使用線性 Diff運算符之情形，DifO及Dif 1較佳為予以線性測量，並表示在中心像素與頭間之差，並且連同線性Diff測量向量之尾像素。在例如c為中心像素之灰值，以及b及a 為頭及尾像素之灰值之情形，較佳為由下式示DifO及 DifO ：Gradient is the input gradient value; and EdgeFactor (NarrowArea) is a factor, usually in the range of 0 and 1, and is selected to give less weight to the gradient near the edge; EdgeFactor (NarrowArea) is preferably near the edge, such as 0.8 value. It calculates the gradient as explained above, analyzes, classifies, and bounds ed (step 470). Please refer to FIG. 26, which is a part of step 470 in FIG. 25. A simplified process diagram of a preferred implementation includes analysis and classification. In Fig. 26, DifO and Dif1 are preferably as described above with reference to step 466. Please note that, as explained above, in the case of using the linear Diff operator, DifO and Dif 1 are preferably measured linearly and represent the difference between the center pixel and the head, together with the tail pixels of the linear Diff measurement vector. In the case where, for example, c is the gray value of the center pixel, and b and a are the gray values of the head and tail pixels, DifO and DifO are preferably represented by the following formula:

DifO^c-b ;以及 -84- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐)DifO ^ c-b; and -84- This paper size applies to China National Standard (CNS) A4 (210 X 297 mm)

裝訂Binding

536919 A7536919 A7

Difl=a-c 〇圖26之方法為自可明白。請回頭參照圖24,在步驟梢然後應用墟定準壚，_私θ 依據預又為自Θ像之先前分析導得，應用度及最大值/最小俏磕令妒社、a时 > 丨限主梯取」值確疋，軚佳為將每一梯度或小值報告等級如下：又值/取〇無報告 1 弱報告 2 中間報告 3 強報告步騾47〇之輸出，在每一情形較佳為包含單獨之彩色1 及彩色2報告，供局部最小值/最大值缺陷，及供梯度缺陷。彩色1及彩色2報告予以統一（步驟472 )，較佳為應用最強者殘存”規則，一般為由下列邏輯所表示：報告值（彩色1)> =報告值（彩色2)) 報告彩色=彩色1 報告值=報告（彩色1) 報告型式=報告型式（彩色1)Difl = a-c 〇 The method in Figure 26 is self-explanatory. Please refer back to FIG. 24, and then apply the market setting criterion at the end of the step. _Private θ is derived from the previous analysis of the Θ image, and the application degree and the maximum value / minimum value make the jealous society, a time > 丨The value of “Limit to the main ladder” is confirmed. The best grade is to report each gradient or small value as follows: Revalue / Take 0 No report 1 Weak report 2 Intermediate report 3 Strong report Step 47 The output of 47 It is preferable to include separate color 1 and color 2 reports for local minimum / maximum defects, and for gradient defects. The color 1 and color 2 reports are unified (step 472), and it is better to apply the strongest surviving rule, which is generally expressed by the following logic: report value (color 1) > = report value (color 2)) report color = Color 1 report value = report (color 1) report type = report type (color 1)

Else 報告彩色=彩色2 報告值=報告（彩色2 ) 報告型式=報告型式（彩色2) -85- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐) 536919 A7Else report color = color 2 report value = report (color 2) report type = report type (color 2)

步驟472之結果然後予以過濾（步驟474 )，一般為藉分析報告〈3 X 3纟開，計算—加權和，並應用預定之界限；較佳為，在硬體LUT進行此操作。 2 .關於子解析度通道404之步驟：使&像平滑，較佳為利用5 χ 5棋盤中間過濾（步驟 476 ),此步騾平行於在全解析度通遒之步騾464。在子解析度通道，搜尋大缺陷；因而，宜於不管小特色，即使其具有與周圍區域之大變異。因此較佳為使用大核心大小過遽，設計為不管與其近鄰大為不同之像素。較佳為，如在此項技藝所熟知，使用一中間過遽，其中資料予以取樣並確定及使用中間值。為供容易實施，希望在一子樣本，諸如5X5棋盤子樣本諸如下列者，進行中間計算，其中標記X之像素予以取樣，並且標記〇者不予以取樣： ΤThe result of step 472 is then filtered (step 474). Generally, the analysis report <3 X 3 is opened, the weighted sum is calculated, and a predetermined limit is applied; preferably, this operation is performed on a hardware LUT. 2. Steps on sub-resolution channel 404: To smooth the & image, preferably using a 5 x 5 checkerboard intermediate filter (step 476), this step is parallel to step 464 in the full-resolution pass. In the sub-resolution channel, search for large defects; therefore, it is better to ignore small features, even if they have large variations from the surrounding area. Therefore, it is better to use a large core size, and design it to be a pixel that does not differ greatly from its neighbors. Preferably, as is well known in the art, an intermediate pass is used in which the data is sampled and intermediate values are determined and used. For ease of implementation, it is desirable to perform intermediate calculations on a sub-sample, such as a 5X5 chessboard sample, such as the following, where pixels marked X are sampled, and those marked 0 are not sampled: Τ

裝Hold

X 0 X 0 X 0 X 0 X 0X 0 X 0 X 0 X 0 X 0

X 0 X 0 X 應用梯度檢測器（步騾478 );此步驟孚A '也、 7鄉千仃於步驟466 並且除了在材料内使用Sobel檢測器，輕干乂住為類似於步 466。S〇bel檢測器係' 在以上，特別是參照步驟偏予以 -86-X 0 X 0 X Apply a gradient detector (step 478); this step is also performed in step 466 and except that the Sobel detector is used in the material, it is lightly similar to step 466. The Sobel detector is above, especially with reference to the steps.

A7A7

進行梯度正常化（步驟48〇)，驟468之梯度正常化。，、較佳為類似於步、在步驟術，進行類似於在平行步驟470之定限之梯产足限，並且彩色報止予梯度 % 〇丁以統一。在步騾482之彩色報止之統-’較佳為類似於㈣472。 : 度缺陷之報告影像。生匕括未過滤梯應用-大區域結果過滤（步驟484);此步驟平行於步驟 474。在步騾484，較佳為首先應用一類似於以上參照步驟474所說明者之3Χ3過濾。然後，使用3χ3過濾之結果作成較大展開，諸如9 χ 9展開，其然後同樣依據一加權和之定限予以過濾。較佳為，9 X 9展開在過濾後予以子取樣’ 一般為根據下列圖案，而χ表示予以取樣之像素，及0表示不予取樣之像素： 0 0 0 0 0 0 0 0 0 0 X 0 0 X 0 0 X 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 X 0 0 X 0 0 X 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 X 0 0 X 0 0 X 0 0 0 0 0 0 0 0 0 0 -87- 本紙張尺度iiiT®國家標準(CNS) A4規格(210X297公釐) 536919 A7 B7 五、發明説明（） 85 結果然後在有一叢集報告之報告”團”時，任選予以稀釋， -88- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐) 536919Gradient normalization is performed (step 48), and the gradient of step 468 is normalized. , Preferably similar to step, step-by-step technique, similar to the limit of ladder production in parallel step 470, and the color stop is given a gradient of% 〇 to unify. The color stop system-'at step 482 is preferably similar to 472. : Report image of degree defect. Use the unfiltered ladder to apply-Large area result filtering (step 484); this step is parallel to step 474. At step 484, it is preferred to first apply a 3 × 3 filter similar to that described above with reference to step 474. Then, use the result of 3 × 3 filtering to make a larger expansion, such as a 9 × 9 expansion, which is then filtered according to a weighted sum limit. Preferably, the 9 X 9 expansion is sub-sampled after filtering. Generally, it is based on the following pattern, and χ represents pixels that are sampled, and 0 represents pixels that are not sampled: 0 0 0 0 0 0 0 0 0 0 X 0 0 X 0 0 X 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 X 0 0 X 0 0 X 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 X 0 0 X 0 0 X 0 0 0 0 0 0 0 0 0 0 -87- Paper size iiiT® National Standard (CNS) A4 size (210X297 mm) 536919 A7 B7 V. Description of the invention () 85 Results Then, when there is a cluster of report "groups", it can be optionally diluted. -88- This paper size applies Chinese National Standard (CNS) A4 (210X297 mm) 536919

團之邊緣之外部報告不予稀釋，為内部報告予以稀釋’因此保持較小代表性報告之數，藉以節省儲存空間。 3·合併子解析度通道報告及全解析度通道報告。子解析度通道報告及全解析度通道報告予以合併（步騾 468 )。合併係藉下列規則所完成： L取強者殘存"，類似於以上參照步驟472 明者。. 2·在二報告具有相同強度時，使用較精細檢測器者，根據下列順序：局邵最小值/最大值：> 全梯度 > 子梯度在一種代表性實施，在最後報告之每一項目包含一 9位元字，一種代表性結構如下：欄位位元值/解釋報告型式 2 〇子梯度 1全梯度 2局部最大值 3局邵最小值報告強度 2 0無報告 1弱報告 2強報告 3稀釋報告報告彩色1 〇強度取向（彩色i) -89·The external reports at the margin of the group are not diluted, and the internal reports are diluted. Therefore, the number of representative reports is kept small to save storage space. 3. Combine sub-resolution channel report and full-resolution channel report. The sub-resolution channel report and the full-resolution channel report are merged (step 468). Merging is accomplished by the following rules: L takes the strong one remaining ", similar to what was explained above with reference to step 472. 2. When the two reports have the same intensity, the user who uses the finer detector, according to the following order: local minimum / maximum value: > full gradient > The project contains a 9-bit word, and a representative structure is as follows: Field value / interpretation report type 2 0 Sub-gradient 1 Full gradient 2 Local maximum 3 Local minimum report intensity 2 0 No report 1 Weak report 2 Strong Report 3 Dilution Report Report Color 10 Strength Orientation (Color i) -89 ·

536919536919

五、發明説明（ 87 寬/窄 1色取向（彩色2 ) 〇寬區域（在材料内）報告材料 1窄區域（靠近材料邊緣）〇-7 (依據彩色分類）渉態骨架產現請參照圖27 、其為操作圖1之線及空間骨架子單元 '種較佳方法之簡化流程圖示。圖27之方法包含一：：方！，腐蝕方法通常在該項技藝為熟知者。線及空 ^子單元230較佳為接收二進制化及[el產生子單元所產生，較佳為自單色影像資料導得之二進制cel作為輸入。圖27之步驟較佳為包含下列：生骨架產生之影像經歷腐蝕，在預定之χ及γ掃描方向橫向移動（步驟49G)，諸如例如左至右，在X方向沿每一列，繼之頂至底在Y方向。口進行腐蝕，供每一像素，一般為根據予以腐蝕之像素之4或8最靠近近鄰（步驟492)。現請另外參照圖“A及 28Β，其為一陣列像素，在部份腐姓前後一間化畫面圖不，用於瞭解圖27之方法。在圖2δΑ中，γ 、 τ 一予以腐蝕 496，標記為a4之像素，可根據下列任_之像素值予、蝕：以腐四水平及垂直近鄰498，標記為al，a3，或及a7 ; 90- 本紙張尺度適用中國國家標準(CNS) A4規格(21QX297公爱) 536919 A7 B7V. Description of the invention (87 wide / narrow 1-color orientation (color 2)) 0 wide area (within the material) report material 1 narrow area (close to the edge of the material) 0-7 (according to color classification) 渉 state skeleton production please refer to the figure 27. It is a simplified flow diagram of a better method of operating the line and space skeleton sub-units of FIG. 1. The method of FIG. 27 includes one: Fang !, and the corrosion method is generally known in the art. Line and space ^ The sub-unit 230 preferably receives the binarization and [el generated by the sub-unit, preferably a binary cel derived from monochrome image data as input. The steps in FIG. 27 preferably include the following: The image generated by the raw skeleton After etching, it moves laterally in the predetermined χ and γ scanning directions (step 49G), such as, for example, left to right, along each column in the X direction, followed by top to bottom in the Y direction. Etching is performed for each pixel, generally The closest neighbor is based on 4 or 8 of the etched pixels (step 492). Now please refer to the drawings "A and 28B, which are an array of pixels. A picture is not shown before and after some rot names, for understanding Method of Figure 27. In Figure In 2δΑ, γ and τ are etched 496, and the pixels labeled a4 can be etched according to the following pixel values: rot four horizontal and vertical neighbors 498, labeled al, a3, or a7; 90- This paper size applies to China National Standard (CNS) A4 specifications (21QX297 public love) 536919 A7 B7

88 八最靠近近鄰包含四水平及垂直近鄰496，以及四對角近鄰500 ’標記為aO，a2，a6，及a8。選擇那些最靠近近鄰，可根據予以進叮又分析而變化。精於此項技藝者將會察知，依據四最靠腐蝕，利於主水平及垂直軸線，而依據八最靠近近鄰之腐蝕，也允許沿對角之腐蝕。基本上，為了允許更均勻腐姓，較佳為使用四及使用八最靠近近鄰腐蝕，但在實際上可使用四或八最靠近近鄰之一。在任一情形，腐蝕較佳2 自取為有1之值之黑像素，至取為有〇之值之白像素，請予察知，腐蝕自白至黑，及/或自以上所述者指定相反值至白及黑像素也為可能。較佳腐蚀規則如下·· 供四近鄰’如果四水平及垂直近鄰498之任何者有〇之值’設定中心像素496至0 ;以及供八近鄰，如果四水平及垂直近鄰498之任何者，或對角近鄰500之任何者有〇之值，設定中心像素496至〇。在本發明，連接性較佳為但不一定在腐蝕時予以保留；亦即，連接黑像素應該持續連接，並且不”間斷·，，即使在腐姓後。如果連接性將行予以保留，應用連接性狀況（步驟494 )，一般為包含Euler狀況，其在該項技藝為熟知者。如果在腐蝕前後保留Euler狀況所指示之和或分數’ Euler狀況為幾何狀況，其意為保留連接性。要保留連接性，如果達反Euler狀況，不進行腐蝕。 -91 - 本紙張尺^^冢標準(CNS) A4規格(寧挪公爱) 裝訂The 88 nearest neighbors include four horizontal and vertical neighbors 496, and four diagonal neighbors 500 'labeled as aO, a2, a6, and a8. Choosing the closest neighbors can change based on analysis and analysis. Those skilled in this art will know that the four closest corrosions are good for the main horizontal and vertical axes, and the eight closest corrosions are also allowed to corrode diagonally. Basically, in order to allow a more uniform family name, it is preferable to use four and eight nearest neighbors, but in practice one of four or eight closest neighbors may be used. In either case, corrosion is better 2 from black pixels with a value of 1 to white pixels with a value of 0. Please be aware that corrosion from white to black, and / or specify the opposite value from the above White and black pixels are also possible. The preferred corrosion rules are as follows: · for the four nearest neighbors 'if any of the four horizontal and vertical neighbors 498 have a value of 0' set the center pixel 496 to 0; Any of the diagonal neighbors 500 has a value of 0, and the center pixel is set to 496 to 0. In the present invention, the connectivity is preferably but not necessarily preserved during corrosion; that is, the connected black pixels should be continuously connected and not "interrupted", even after the rotten name. If connectivity will be preserved, apply The connectivity condition (step 494) generally includes the Euler condition, which is well known in the art. If the sum or score indicated by the Euler condition is preserved before and after corrosion, the Euler condition is a geometric condition, which means that connectivity is preserved. To maintain connectivity, do not corrode if it is in anti-Euler condition. -91-This paper ruler ^^ Tsukazaki Standard (CNS) A4 Specification (Ningnuo Public Love) Binding

536919 A7 B7 五、發明説明（β。）〇y 請參照圖28B，已予腐蝕502之像素予以指示為pO， pi，p2，p3。如果滿足以上關於步騾492所述之狀況，並也滿足Euler狀況，予以腐蝕之像素496較佳為僅予以腐蝕。現請另外參照圖29，其為Euler狀況之簡化畫面圖示。在圖29中，將行加及減許多圖案504之發生數如在圖29中所指示，俾獲得在腐蝕前後應該為不變量之和。較佳為，如以上所說明，為供操作效率，在定硬體實施圖 27之方法。請予察知，圖27之方法可進行一或許多次供每一影像，因此進行一個或多個階段之腐I虫。較佳為，所進行腐蝕之階段數，根據予以腐蝕之任何特色之最大厚度而變化，以便在腐蝕結束時，獲得一像素寬度之骨架。每一階段可對4或8近鄰進行，並且保留或不保留連接性，請予察知，在有些腐蝕使用4近鄰，而在其他腐蝕使用8近鄰，可有助於獲得腐蝕之各向同性。現請參照圖30A-30C，其為簡化畫面圖示，使用於暸解圖1之形態子單元240之操作之一種較佳方法。形態子單元240較佳為自線及空間骨架子單元230接收輸入，及自材料識別接收均勻彩色粒子數之影像。在圖30A中，示一影像在腐蝕506前之實例。在圖30B 中，示影像之黑色508在圖30A之腐蝕506前，其腐蝕結果之實例。在圖30C中，示影像之白色510在圖30A 之腐蝕506前，其腐蝕結果之實例。 -92- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐)536919 A7 B7 V. Description of the invention (β.) 〇y Please refer to FIG. 28B. The pixels that have been etched 502 are indicated as pO, pi, p2, p3. If the condition described above with respect to step 492 is satisfied, and the Euler condition is also satisfied, the pixel 496 to be etched is preferably only etched. Please also refer to FIG. 29, which is a simplified screen diagram of the Euler condition. In Fig. 29, the number of occurrences of adding and subtracting a number of patterns 504 as indicated in Fig. 29, 俾 obtains a sum that should be invariant before and after etching. Preferably, as explained above, the method of FIG. 27 is performed on the hardware for operational efficiency. Please note that the method of FIG. 27 can be performed one or more times for each image, and therefore, one or more stages of rotten worms are performed. Preferably, the number of stages of etching is changed according to the maximum thickness of any feature to be etched, so that at the end of the etch, a one-pixel-wide skeleton is obtained. Each stage can be performed on 4 or 8 neighbors, with or without continuity. Please be aware that using 4 neighbors for some corrosion and 8 neighbors for other corrosion can help to obtain isotropic corrosion. Please refer to FIGS. 30A-30C, which is a better method for understanding the operation of the morphology sub-unit 240 of FIG. The morphology sub-unit 240 preferably receives input from the line and space skeleton sub-unit 230, and receives images of uniformly colored particles from material identification. In FIG. 30A, an example of an image before etching 506 is shown. In Fig. 30B, an example of the result of the etching of the black 508 of the image before the etching 506 of Fig. 30A is shown. In FIG. 30C, an example of the corrosion result of the white 510 of the image before the corrosion 506 of FIG. 30A is shown. -92- This paper size applies to China National Standard (CNS) A4 (210X297 mm)

裝訂 t 五、發明説明（在圖30B及30C中，示將行件之與纟U it · π ^ 了以私挪之代表性形態事件（只例如下·開口端512 ;紝 516。 ϋ ^ 514 ;以及一區段結點現請參照圖31A_31C，其為運篡^ _ m、a ★ 异付 < 間化畫面圖示，使用於瞭解圖30A-30C。在本發明乏山成斗寸 —種較佳實施例，如在此項技藝所熟知，在圖31A_31C申 —、卜 r所不’也稱為核心之運异付之貫例，可應用於圖3〇B之像或圖30C之影像，以檢測形態事件。在圖3 ^ A中干問 . 〒不開口端核心運算符518，在圖31B中示結點運算符52〇。左在圖31C中，示一區段結點運算符522，其可使用供干么匕 J便用供π被白色所包圍之”黑色，，島。容易察知，雖然根據本發明之一種較佳實施例，關於 -黑色及白色影像，或二進制影像所示，源自二進制化及 cel產生子單元2〇〇，但在圖中所示3〇b及3〇c之腐蝕二進制影像之部位及特色，係相關於均勾彩色粒子數。以此方式，特定特色，諸如512 ;結點5M ;以及區段結點 516，可相關於均勻彩色粒子數，並且如以上所說明，另相關於各別均勻彩色粒子數所表示之材料及材料組合。現請參照圖32，其為彩色形態分析影像之一種較佳方法之簡化流程圖圖示。為供說明簡單之目的，圖32之方法在二進制形態分析影像之說明後予以說明，請予察知，圖32之方法可例如至少部份在圖1之clrp單元120實施’其中彩色資訊為現有。請予察知，圖32之方法通常 -93- 91 536919 五、發明説明（Binding t V. Description of the invention (In Figs. 30B and 30C, it is shown that the combination of the paperwork with U U · π ^ is a representative event of private movement (for example, only the lower open end 512; 纴 516. ϋ ^ 514 ; And a section node, please refer to FIG. 31A_31C, which is an operation ^ _ m, a. A preferred embodiment, as is well known in the art, is shown in Figs. 31A-31C, which is also referred to as the core of the different examples of payment, which can be applied to the image in Fig. 30B or the image in Fig. 30C. Image to detect morphological events. Ask in Figure 3 ^ A. 〒The open-end core operator 518 is shown in Figure 31B as the node operator 52. The left is shown in Figure 31C as a segment node calculation. Symbol 522, which can be used for the purpose of "Dark, surrounded by white", "black," island. It is easy to know, although according to a preferred embodiment of the present invention, about-black and white images, or binary images As shown, it is derived from the binarization and cel generation subunit 200, but the corrosion binaries of 30b and 30c are shown in the figure. The location and characteristics of the image are related to the number of uniformly colored particles. In this way, specific characteristics, such as 512; node 5M; and segment node 516, can be related to the number of uniformly colored particles, and as explained above, Also related to the materials and material combinations represented by the number of individual uniformly colored particles. Now refer to FIG. 32, which is a simplified flowchart illustration of a better method for color morphological analysis of images. For simplicity, FIG. 32 The method is explained after the description of the binary morphological analysis image. Please note that the method of FIG. 32 can be implemented at least in part, for example, in the clrp unit 120 of FIG. 1 where the color information is available. Please note that the method of FIG. 32 is generally -93- 91 536919 V. Description of the invention (

可用於影像之彩色形態分析，並可因此也在圖！之CRLP 單元⑽之特^範圍外，使料彩色影像分析之其他方面。精於此項技藝者將會察知，_ 32之方法可❹供分析影像之任何適當光學特徵’特別是s 32之方Can be used for color morphological analysis of images, and can therefore also be drawn! Beyond the special range of the CRLP unit, other aspects of material color image analysis are enabled. Those skilled in this art will know that the _ 32 method can be used to analyze any appropriate optical characteristics of the image ’, especially the s 32 method.

形態分析。 W 圖32之方法較佳為包含下列步騾：使用彩色方法產生彩色材料影像（步驟523 )，一般為如二上=彩色處理所說明1色材料影像較佳為由材料識力J子早兀1 7 0所產生’並且每—材料係相關於一均勻彩色匕子數。容易察知’產生彩色材料影像—般顯著減低在 Θ色影像資料中相對於粒子數之數之不同彩色粒子數之數。產之之進較佳為在材料之間造成—間隙（步驟My，_般為藉指足-人工彩色值至沿材料間之每一邊界，在每一區域二素，俾使材料區域分開供另外處理。自步驟似之輸出生二進制影像（步驟525 )，一般為使用在此項技藝熟知万法，諸如足限。彩色影像可代之為藉在此項技藝已知任何習知方法，轉換至_灰度級單色影像，並提供至二制化及cel產生子單元鳩，自其產生二進制影像。驟 2步驟525之輸出進行骨架方法（步驟526);較佳為使 ^ 之方法。在輸出步驟526識別形態事件（步 527)，較佳為如以上參照圖3〇A_3〇c所說明。 94- X 297公釐) 本纸張尺度適用中 536919 A7 B7Morphological analysis. The method of FIG. 32 preferably includes the following steps: Use a color method to generate a color material image (step 523), which is generally as described in the two above = color processing. The one-color material image is preferably made of material awareness. 1 7 0 'and each-material is related to a uniform number of colored daggers. It is easy to know that 'producing color material images—generally significantly reduces the number of different colored particles relative to the number of particles in the Θ color image data. The advancement of production is preferably caused by gaps between materials (step My, _ is generally by means of finger-artificial color values along each boundary between materials, two elements in each area, so that the material areas are separated for supply Other processing. The output binary image (step 525) similar to the step is generally used in this technique, such as the limit. Color image can be replaced by any known method by this technique. To _ gray-scale monochrome images, and provide to binarization and cel to generate sub-unit doves, from which binary images are generated. The output of step 2 step 525 is performed by the skeleton method (step 526); The morphological event is identified in the output step 526 (step 527), preferably as described above with reference to Fig. 30A_30c. 94- X 297 mm) 536919 A7 B7

92 五、發明説明（然後例如在根據以上步驟523所產生之下面材料影像之對應地點之彩色，將特色”著色”，藉以使形態事件與均勻彩色粒子數相關（步騾528 )。根據本發明之一種較佳實施例，著色步騾係藉重疊腐蝕骨架影像及材料影像予以實施，以便在骨架影像之特色根據在重疊材料影像之均勻彩色粒子數之彩色之一接收彩色。因此，產生一彩色形態影像，包含一形態圖，具有來自指定至特色之均句彩色粒子數，或材料之彩色。 _凹隙及凸刼韶虛搜現清參照圖33A及33B，其一起包含操作圖丨之凹隙及凸起邰子單兀220之一種較佳方法之簡化流程圖示。，，凹隙”及"凸起部”二詞指在邊緣之小缺陷；凹隙為伸入邊緣之缺陷，而凸起部為邊緣自其適當地點向外延伸之缺陷。圖1之凹隙及凸起部子單元22〇較佳為可操作識別凹隙及凸起部，並自其影像予以分析之物體之法線特色，諸如例如角部辨別凹隙及凸起部。請予察知，檢測凹隙及凸起邵，在檢查諸如在本文中所說明之物體，其中存在很多有直邊緣之小特色，及其中缺陷諸如凹隙及凸起部為相當常見，並且一般為不合宜時，特別具有意義。現請另外參照圖35,其為將行予以分析之影像，其一祁k之簡化畫面圖示，可用於瞭解圖33A及33B之方法。圖35之影像包括一凹隙8〇〇，一凸起部81〇，一角 •95-92 V. Description of the invention (Then, for example, in the color of the corresponding location of the following material image generated according to the above step 523, the feature is "colored", so that the morphological event is related to the number of uniformly colored particles (step 528). According to the present invention In a preferred embodiment, the coloring step is implemented by overlapping the etched skeleton image and the material image, so that the characteristic of the skeleton image is received in accordance with one of the colors of the uniform number of colored particles in the overlapping material image. Therefore, a color is generated. The morphological image includes a morphological map, with the number of uniform particles from the specified to the characteristic, or the color of the material. Simplified flow diagram of a better method for gaps and raised rafters 220. ,, "Dent" and "Rent" refers to small defects at the edges; dimples are defects that protrude into the edges, The raised portion is a defect in which the edge extends outward from its proper place. The recessed and raised portion subunit 22 of FIG. 1 is preferably operable to identify the recessed and raised portion, and Normal characteristics of the object to be analyzed, such as discerning recesses and protrusions at the corners, for example, please be aware that detecting recesses and protrusions, inspecting objects such as those described in this article, many of which are straight The small characteristics of the edges, and the defects such as recesses and protrusions are quite common, and they are generally not suitable when they are not suitable. Please refer to FIG. 35, which is an image to be analyzed. The simplified screen icon can be used to understand the method of Figures 33A and 33B. The image of Figure 35 includes a recess 800, a raised portion 810, and a corner • 95-

536919 A7 B7 五、發明説明（） 93 部 in 815及二角部out 820，均在沿邊緣830。在本發明之一種較佳實施例，凹隙及凸起部檢測係依據分析界定不同材料間之邊界之邊緣輪廓，較佳為藉測量相鄰輪廓元件當中之偏差之孔徑，並藉在沿邊緣830行進時，在一至邊緣 830之法線之角度變化，將可疑缺陷分類。請予察知，凹隙及凸起部具有法線之角度變化之相反分布，因而凹隙 800可具特徵為有一種’’in-out-in”變化，同時凸起部 810有一種’'out-in-out··角度變化。在另一方面，角部-in 815有一種nin-in_in"角度變化，而角部out 820 有一種’’out”角度變化。因此請予察知，測量及分析法線角度改變之適當方法，應該能在凹隙，凸起部，與法線特色諸如角部之間辨別。圖33A及33B之方法較佳為包括下列步騾：請予察知，為操作效率起見，步驟840及930 (均在以下說明）較佳為予以並行實施，每一步驟840及930接收將行予以分析之cel作為輸入。在一影像進行間斷檢測，供每一予以分析之cel，一般為供在影像之所有cel (步驟840)。步驟840較佳為包含下列步驟 850，860，870，880，890，900，及 910 : 在一予以分析之cell，測量cel長度及cel指向之法線角度（步驟850 )。請回頭參照圖9-11，請予察知，較佳為 -96- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐） 536919 A7 B7 五、發明説明（） 94 僅有一小預定數之可能cel，並因此可較佳為自LUT確定 cel長度及cel角度。然後較佳為進行下列步騾 870，880，890，900，及 910，供予以分析之cel之每一近鄰cel (步騾860)。請予察知，每一 cel僅有4近鄰，由於，請再次回頭參照圖9-11，請予察知，cel僅可在4近鄰之一之方向並且不在4” 角部π方向之一繼續。檢查近鄰cel之長度，以確保近鄰cel長度為高於一預先界定之界限，諸如例如一 1/4像素之界限（步騾870)。如果近鄰c e 1長度不高於預先界定之界限，繼續處理次一近鄰cel (步驟880 )。如果近鄰cel長度高於預先界定之界限，確定近鄰cel 指向之法線角度，較佳為使用一類似於以上參照步驟850 所說明者之方法，並計算間斷角度（步騾890 )。現請另外參照圖34，其為簡化畫面圖示，有助於暸解圖33A之步驟890。圖34示一 cel 891以及近鄰cel 892及893。分別示至cel 892及893之法線894及895。間斷角度897於是為cel 891之間斷角度。檢查間斷角度是否大於一預定之界限，諸如例如一大約22度之界限（步驟900 )。如果間斷角度不大於預定之界限，繼續處理以上所提及之步驟880。如果間斷角度大於預定之界限，標記一間斷供予以分 -97- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) 參裝訂536919 A7 B7 V. Description of the invention () 93 sections in 815 and two corner sections out 820 are all along the edge 830. In a preferred embodiment of the present invention, the detection of the recesses and protrusions is based on the analysis to define the edge contours of the boundaries between different materials, preferably by measuring the aperture of the deviation between adjacent contour elements, and borrowing along the edges As 830 travels, the angle changes from the normal to the edge 830 to classify suspicious defects. Please note that the recesses and protrusions have opposite distributions of normal angle changes, so the recesses 800 can be characterized as having an `` in-out-in '' change, while the protrusions 810 have an `` out -in-out ·· Angle change. On the other hand, corner-in 815 has a nin-in_in " angle change, and corner out 820 has a `` out "angle change. Therefore, please be aware that the proper method of measuring and analyzing the normal angle change should be able to distinguish between the recesses, raised portions, and normal features such as corners. The method of FIGS. 33A and 33B preferably includes the following steps: Please note that, for operational efficiency, steps 840 and 930 (both described below) are preferably implemented in parallel, and each step 840 and 930 will perform Analyze the cel as input. Intermittent detection is performed on an image, and each cel for analysis is generally all cels provided for the image (step 840). Step 840 preferably includes the following steps 850, 860, 870, 880, 890, 900, and 910: In a cell to be analyzed, measure the cel length and the normal angle of the cel direction (step 850). Please refer back to Figure 9-11, please be aware, preferably -96- This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) 536919 A7 B7 V. Description of the invention () 94 There is only a small The predetermined number of possible cels, and therefore it may be preferable to determine the cel length and cel angle from the LUT. It is then preferable to perform the following steps 870, 880, 890, 900, and 910 for each neighbor cel of the cel to be analyzed (step 860). Please be aware that each cel has only 4 nearest neighbors. Since, please refer to Figure 9-11 again. Please be aware that cel can only continue in one of the 4 nearest neighbors and not in one of the 4 "corner π directions. Check The length of the neighbor cel to ensure that the length of the neighbor cel is higher than a pre-defined limit, such as, for example, a 1/4 pixel limit (step 870). If the length of the neighbor ce 1 is not higher than the pre-defined limit, continue processing times A neighbor cel (step 880). If the length of the neighbor cel is higher than a pre-defined limit, determine the normal angle at which the neighbor cel points, preferably using a method similar to that described above with reference to step 850, and calculate the discontinuity angle ( Step 890). Now please refer to FIG. 34, which is a simplified illustration of the screen and helps to understand step 890 of FIG. 33A. FIG. 34 shows a cel 891 and its neighbors cel 892 and 893. It is shown to cel 892 and 893 respectively. Normals 894 and 895. The discontinuity angle 897 is then the discontinuity angle of cel 891. Check if the discontinuity angle is greater than a predetermined limit, such as, for example, a limit of approximately 22 degrees (step 900). If the discontinuity angle is not greater than the predetermined limit Limit, continue to process step 880 mentioned above. If the discontinuity angle is greater than the predetermined limit, mark a discontinuity for division -97- This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm).

536919 A7 " ______________B7 五、發明説明（1 一 95 析之cel (步驟910)，一般為在一適當圖。為了確保繞在步驟840予以分析之cel之cd之部位予以標記為有問豸，造成一間斷圖展開，因此使間斷報告有效擴展；T即，二相鄰像素，包含二像素，其中予以：析又cel繼續，也標記為有一間斷（步驟92〇)。在擴展間斷報告時，較佳為不重寫記錄供一相鄰像素之先前報告，俾防止一報告之間斷被一相鄰間斷之擴展所重寫。然後抽取間斷報告（步騾925 )並傳遞至以下所說明之步驟 980。較佳為並行於圖33A之上述步驟，在一影像進行特色特徵說明，供每一予以分析之cd，一般為供在影像之所有(步騾930)。步驟93〇較佳為包含下列步驟94q , 950，960，及 970 : 確足cel之一中點（步驟94〇 )。請予察知，如以上所說明’由於在每一虛擬cel像素僅有一有限數之可能cd，故可較佳及最有效率經由LUT求出cel中點。在一繞c e 1予以分析之展開内，一般為在中心在予以分析之cel周圍之5 X 5陣列之虛擬cel像素内，確定二邊緣cel，供目前予以分析之cei為其一部份之特色（步驟 95〇 )。請予察知’供實施效率之部位，將宜於自邊緣cel 經由LUT確定特色深度及孔徑，但一種使用5 X 5陣列之直截了當LUT實施，將需要25位元之LUT位址，或 -98-536919 A7 " ______________B7 V. Description of the invention (1-95 analysis of cel (step 910) is generally a suitable figure. To ensure that the part of the cd around the cel analyzed in step 840 is marked as questionable, causing An interrupted graph is expanded, so the interrupted report is effectively expanded; T is, two adjacent pixels, including two pixels, of which: analysis and cel continue, also marked as an interrupted (step 92). When the interrupted report is expanded, the It is better not to rewrite the previous report for an adjacent pixel, to prevent a report break from being overwritten by an adjacent break extension. Then extract the break report (step 925) and pass to step 980 described below. It is preferable to perform the above-mentioned steps in parallel with FIG. 33A, and to describe the characteristic features of an image. Each cd for analysis is generally provided for all the images (step 930). Step 930 preferably includes the following steps. 94q, 950, 960, and 970: exactly one midpoint of cel (step 94). Please note that as explained above, 'because there is only a finite number of possible cds in each virtual cel pixel, it may be better The most efficient method is to find the midpoint of the cel through the LUT. In the unfolded analysis of a circle of ce 1, it is generally to determine the two-edge cel in the virtual cel pixel of the 5 X 5 array around the cel being analyzed, for the current The cei to be analyzed is part of its characteristics (step 95). Please be aware of the 'for implementation efficiency, it will be suitable to determine the characteristic depth and aperture from the edge cel through the LUT, but a straightforward use of a 5 X 5 array LUT implementation will require a 25-bit LUT address, or -98-

本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) J 536919 A7 B7 五、發明説明（ ) 96 項目，其被視為在實際上太難以處理之項目數。供在步騾950 求出邊緣cel之一種較佳方法包含下列。現請另外參照圖 36，其為使用於暸解圖33B之步騾950 之一陣列cel虛擬像素之簡化畫面圖示。在圖36之實例，一目前予以分析之cel或中心cel 951為已知。一第一肩角cel 952及第二肩角cel 953有待予以確定。較佳為，首先在九中心虛擬cel像素954，包含中心cel 951之虛擬像素及其八一階近鄰進行3X3連接性分析。連接性分析之目的為指定一型式至予以檢查之輪廓，一般為根據下列：連接性型式連接性碼主要對角 0 次要對角 1 垂直 2 水平 3 上左 4 上右 5 左下 6 右下 7 無連接性 8 有缺陷輪廊 15 (OFx) 現請參照圖37A及37B，其包含輪廓之實例之簡化畫 -99- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) 536919 A7 B7 五、發明説明（） 97 面圖示，有助於瞭解上表。圖37A包含一有缺陷輪廓之實例，在發現超過一輪廓時，其通常予以確定為存在。圖 37B包含一上左輪廓之實例，類似於圖36中所示者。較佳為，進行分析16邊緣虛擬cel像素955之連接之組份，諸如連接之cel。請回頭參照圖36之實例，第一及第二輪廓係例如見於16邊緣虛擬cel像素955。然後較佳為應用下列邏輯，其中η表示見於16邊緣虛擬cel像素 955之性質不同輪廓之數： η 操作〇或1發現未連接輪廓，不報告 2或3 如以下所說明進行 >3 報告有缺陷輪廓（空間達反或凹隙/凸起部）其次，先前所發現輪廓之二輪廓，予以選擇為最可能予以連接在一起，較佳為依據距離測量，以便較佳為使用最靠近位於九中心虛擬cel像素954之中心輪廓之二輪廓。較佳為，根據歐幾里德（Euclidean)距離，發現彼此最靠近之二所選擇輪廓之二cel，並報告作為可疑輪廓之肩角。例如，藉上述方法確定圖36中之第一肩角cel 952及第二肩角cel 953。請回頭參照圖33B，依據在步驟950所確定之二邊緣 cel，及在步驟940所確定之中點，在本發明之一種較佳 -100- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐）This paper size applies to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) J 536919 A7 B7 V. Description of invention () 96 items, which are regarded as the number of items that are too difficult to handle in practice. A preferred method for obtaining the edge cel at step 950 includes the following. Please refer to FIG. 36 additionally, which is a simplified screen diagram for understanding the array of cel virtual pixels in step 950 of FIG. 33B. In the example of Figure 36, a cel or central cel 951 that is currently being analyzed is known. A first shoulder angle cel 952 and a second shoulder angle cel 953 have yet to be determined. Preferably, a 3X3 connectivity analysis is first performed on the nine center virtual cel pixels 954, the virtual pixels including the center cel 951, and their eighth-order neighbors. The purpose of connectivity analysis is to specify a type to the contour to be checked, which is generally based on the following: Connectivity type Connectivity code Major diagonal 0 Minor diagonal 1 Vertical 2 Horizontal 3 Upper left 4 Upper right 5 Lower left 6 Lower right 7 No connectivity 8 Defective contour 15 (OFx) Please refer to Figures 37A and 37B for simplified drawings including examples of outlines -99- This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) 536919 A7 B7 V. Description of the invention () 97 surface diagrams are helpful to understand the above table. Figure 37A contains an example of a defective contour, which is usually determined to be present when more than one contour is found. FIG. 37B contains an example of the upper left outline, similar to that shown in FIG. 36. Preferably, the connected components of the 16-edge virtual cel pixel 955 are analyzed, such as the connected cel. Referring back to the example of FIG. 36, the first and second contours are, for example, seen at 16-edge virtual cel pixels 955. Then it is better to apply the following logic, where η represents the number of different contours of the nature seen in the 16-edge virtual cel pixel 955: η operation 0 or 1 found unconnected contour, do not report 2 or 3 Perform as described below> 3 report with Defect contour (inverse space or recess / protrusion) Secondly, the two contours of the previously found contours are selected as the most likely to be connected together, preferably based on distance measurement, so it is better to use the closest The center outline cel pixel 954 is the second outline of the center outline. Preferably, based on the Euclidean distance, two of the selected contours which are closest to each other are found to be cel, and the shoulder angle which is a suspicious contour is reported. For example, the first shoulder angle cel 952 and the second shoulder angle cel 953 in FIG. 36 are determined by the above method. Please refer back to FIG. 33B. According to the second edge cel determined in step 950 and the midpoint determined in step 940, in a preferred embodiment of the present invention, the paper size is -100-the Chinese National Standard (CNS) A4 specification ( 210 X 297 mm)

裝訂Binding

536919 A7 B7 五、發明説明（） 98 實施例，三特徵說明點予以變換，以供另外處理之簡單性 (步驟960)。請予察知，在本發明之另一實施例，步騾 960可予以省略，由於存在步騾960主要有助於有效率計算。請予察知，中心cel之特徵可為4位元之資訊，而二邊緣cel可各予以特徵為藉4位元之地點資訊及4位元之 cel位址資訊，總共20位元之資訊。進行剛性變換至三點，使中心cel中點移位至中心cel虛擬像素之中心，或至地點已知為優先之另一適當固定點，便需要較少資料。一般為，可省略4位元之資訊，因此僅需要16位元之資訊。現在依據在步騾960所變換之變換點，確定特色深度及孔徑（步驟970)。較佳為，深度及孔徑經由LUT，依據以上參照步驟960所說明之16位元之資訊，依據自中心cel 虛擬像素之右下角部至一連接第一及最後cel之線之深度測量予以確定。現請另外參照圖38，其為一凸起部之代表性實例之簡化畫面圖示，示其深度971及孔徑972。現請參照圖39A，其為一陣列虛擬cel像素之簡化畫面圖示，使用於暸解圖33B之步驟980。圖39A之陣列包含一已檢查之輪廓981，有一中心虛擬cel像素982，一第一肩角983，及一最後肩角984。請另外參照圖39B，其為圖39A之虛擬cel像素陣列之簡化畫面圖示，分別示 "in”及noutn間斷報告之地點，由’Τ’及所標示，與所檢 -101 - 本紙張尺度適用中國國家標準(CNS) Α4規格(210X 297公釐） 536919 A7 B7 五、發明説明（查輪廓9 8 1之分析關聯。較佳為，僅分別對應於中心像素982，第一肩角983，及最後肩角984之間斷報告 985，986，及987，予以輸入以下所說明之步騾980。步騾980較佳為自步騾925及自步驟930均接收輸出，均為以上所說明。較佳為，在步騾980，依據自步騾925 及930所接收之輸入，實施一般為包含LUT之判定表，藉以確定一結果。較佳為自適當之LUT確定步騾980之結果。一般為，至LUT之輸入包括輪廓深度及孔徑；以及在輪廓之中心及在肩角所記錄之間斷，其也分別稱作中心，第一，及最後間斷。LUT之結果一般為包含下列之一：凹隙；凸起部；角部；無報告。LUT實施之適當方法，一般為包含一種蠻力判定方法，包含導致判定之一連串邏輯操作。較佳邏輯操作可說明於試探諸如例如： -一間斷順序in-in-in指示角部，但如果其深度大為超過深度界限，諸如例如多於深度界限二倍，可予以報告為凸起部；一間斷順序in-out-in指示一凹隙，並且如果超過一最小凹隙深度界限，將予以報告為一凹隙；以及唯有其深度為足夠大及其寬度為足夠窄，才報告none-out-in之間斷順序。一般為，深度界限為應用或予以檢查之物體之型式所特定，並且一般為可在一像素之3/4與2或3像素之間變 -102- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐)536919 A7 B7 V. Description of the invention () 98 embodiment, three feature description points are transformed for the simplicity of additional processing (step 960). Please note that in another embodiment of the present invention, step 960 can be omitted, because the existence of step 960 is mainly helpful for efficient calculation. Please know that the center cel can be characterized by 4-bit information, and the two edge cels can each be characterized by borrowing 4-bit location information and 4-bit cel address information, a total of 20-bit information. Performing a rigid transformation to three points, shifting the center cel midpoint to the center of the center cel virtual pixel, or to another appropriate fixed point where the location is known to be a priority, requires less data. Generally, 4-bit information can be omitted, so only 16-bit information is required. The characteristic depth and aperture are now determined based on the transformation points transformed in step 960 (step 970). Preferably, the depth and the aperture are determined via the LUT, based on the 16-bit information described above with reference to step 960, based on the depth measurement from the lower right corner of the central cel virtual pixel to a line connecting the first and last cel. Please refer additionally to FIG. 38, which is a simplified screen illustration of a representative example of a raised portion, showing its depth 971 and aperture 972. Please refer to FIG. 39A, which is a simplified screen diagram of an array of virtual cel pixels, used for understanding step 980 of FIG. 33B. The array of FIG. 39A includes a checked contour 981, a center virtual cel pixel 982, a first shoulder angle 983, and a last shoulder angle 984. Please also refer to FIG. 39B, which is a simplified screen icon of the virtual cel pixel array of FIG. 39A, showing the locations of " in " The scale is applicable to the Chinese National Standard (CNS) A4 specification (210X 297 mm) 536919 A7 B7 V. Description of the invention (Analytical correlation of checking contour 9 8 1. Preferably, it only corresponds to the center pixel 982 and the first shoulder angle 983 , And the last shoulder angle 984 break reports 985, 986, and 987, and enter the step 980 described below. Step 980 preferably receives output from step 925 and step 930, all of which are described above. Preferably, at step 980, based on the inputs received from steps 925 and 930, a decision table that generally includes an LUT is implemented to determine a result. Preferably, the result of step 980 is determined from an appropriate LUT. General For, the input to the LUT includes the contour depth and aperture; and the discontinuities recorded at the center of the contour and at the shoulder angle, which are also called the center, first, and last discontinuities. The results of the LUT generally include one of the following: Recess There is no report. The appropriate method for LUT implementation generally includes a brute force judgment method, including a series of logical operations that lead to the judgment. The better logical operations can be explained in heuristics such as:-an intermittent in-in-in indicator It can be reported as a raised portion if its depth exceeds the depth limit, such as, for example, twice as much as the depth limit; a discontinuous sequence of in-out-in indicates a notch, and if it exceeds a minimum notch depth The boundary will be reported as a notch; and the none-out-in discontinuity sequence will be reported only if its depth is large enough and its width is narrow enough. Generally, the depth limit is the type of object being applied or inspected Specific, and generally variable between 3/4 of a pixel and 2 or 3 pixels -102- This paper size applies to China National Standard (CNS) A4 specifications (210X297 mm)

裝訂Binding

536919 A7 B7 五、發明説明（） 100 化，同時孔徑界限在寬度一般為2.5像素。步騾980之結果較佳為予以報告，結果包含凹隙/凸起部檢測之報告，並可予以後處理以產生報告，一般為包含觸發此種後處理之觸發。如以上所說明之結果，一般可包含：凹隙；凸起部；角部；或無報告。現請參照圖39C及39D，其為可配合圖1之系統使用之方法之簡化流程圖。圖39C之方法包含一種供自彩色影像生產彩色形態圖之方法。圖39D之方法包含一種供在彩色影像確定彩色cel之方法。圖39C及39D之方法按照以上說明而為自可明白。現請參照圖40，其為可使用供暸解本發明之操作之有圖案物品諸如電路，其一種實例之簡化晝面圖示。圖 40 之裝置意為提供一可使用本發明予以分析之物體之特定實例。圖40之實例不意為限制。圖40之裝置包含電路1000，在圖40之實例包含球狀格柵陣列（B G A )基片。如在此項技藝所熟知，BGA 1000 一般為包含：銅導體之部位，被部份透明或半透明塗層諸如焊料掩模1004所覆蓋之銅導體之部位，裸基片1006之部位，被部份透明或半透明焊料掩模1008所覆蓋之基片之部位，塗布金屬鍍層諸如金，例如球1010及電源線 1012之銅導體之部位。另外，鍍敷金之有些部位也可予以覆蓋部份透明或半透明焊料掩模，諸如覆蓋之球 -103- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐）536919 A7 B7 V. Description of the invention () 100, meanwhile, the aperture limit is generally 2.5 pixels in width. The result of step 980 is preferably to be reported, and the result includes a report of the detection of the recesses / protrusions, and can be post-processed to generate a report, which generally includes a trigger to trigger such post-processing. The results as described above can generally include: recesses; protrusions; corners; or no report. Please refer to FIGS. 39C and 39D, which are simplified flowcharts of the method that can be used with the system of FIG. The method of Fig. 39C includes a method for producing a color morphology map from a color image. The method of Fig. 39D includes a method for determining a color cel in a color image. The method of Figs. 39C and 39D is self-explanatory as described above. Reference is now made to Fig. 40, which is a simplified daytime diagram of an example of a patterned article, such as a circuit, that can be used to understand the operation of the present invention. The device of Figure 40 is intended to provide a specific example of an object that can be analyzed using the present invention. The example of Figure 40 is not meant to be limiting. The device of FIG. 40 includes a circuit 1000, and the example of FIG. 40 includes a spherical grid array (B G A) substrate. As is well known in the art, BGA 1000 generally includes: copper conductor parts, parts of copper conductors covered by partially transparent or translucent coatings such as solder mask 1004, parts of bare substrate 1006, and bedding parts. Parts of the substrate covered by the transparent or translucent solder mask 1008 are coated with metal plating such as gold, such as the copper conductors of the ball 1010 and the power cord 1012. In addition, some parts of the plated gold can also be covered with partially transparent or translucent solder masks, such as covered balls -103- This paper size applies to China National Standard (CNS) A4 (210 X 297 mm)

1014。請予察知，在上文所說明夕 — 4 %月义不同材料及材料組合，其不同粒子數，並可產生二進制特色，以表示每一部位。精於此項技藝者也將會察知母一邵位特徵為一不同之辨別彩色粒子數。例如，裸銅及金鍵敷銅之部位為可藉彩色辨別。烊料掩模_般為綠色，及被焊料掩《蓋之銅-般將為淺綠色，被烊料掩模覆蓋 ^鍍金銅將為很淺綠色，以及被焊料掩模覆蓋之基片將為深綠色。在上文所說明本發明之較佳實施例，可識別每一 cel，彩色cel及彩色形態 ’一有圖案物品如BGA，可使用本發明予以分折，句虹八化々^ ^ 士 G括刀析各不同表面缺陷，諸如例如擦痕1016及表面缺陷, 农囬妖1018，例如汙點，凹點，及其他缺陷。發明人等曾發現，在外部處理步驟可能宜於另外處理 BGA之有些部位，不論在該區域是否檢測到缺陷。例如有些部位可由使用者予以界定，及有些部位可與特定特色關聯。因Λ ’例如在特定地點，在若干所考慮之部位，諸如電源線或球之部位，較佳為產生_快照影像，並備供另外之影像處理，不論在該部位是否見有任何特定特色或缺 fe。另外並且代 < 為，在檢測某些特色時，例如一在鍍金銅連接器之開口端，不論位# BGA何處，較佳為產生一快照影像，並備供另外之影像處理，而不論地點，或是否 -104-1014. Please be aware that in the above-mentioned description, the different materials and material combinations of 4% Yueyi, the number of particles, and the binary characteristics can be generated to represent each part. Those skilled in this art will also know that the mother-shao position is a different number of distinguishing colored particles. For example, bare copper and gold-bonded copper areas can be identified by color. The mask is generally green, and the masked copper is generally light green, covered by the mask ^ gold-plated copper will be very light green, and the substrate covered by the solder mask will be dark green. In the preferred embodiment of the present invention described above, each cel, color cel, and color form can be identified. A patterned item such as BGA can be divided by using the present invention. Different types of surface defects such as scratches 1016 and surface defects, Nong Hui Yao 1018, such as stains, pits, and other defects are analyzed. The inventors have found that it may be appropriate to treat some parts of the BGA in an external processing step, regardless of whether a defect is detected in the area. For example, some parts can be defined by the user, and some parts can be associated with specific features. Because Λ ', for example, in a specific place, in a number of places under consideration, such as the power cord or ball, it is better to generate a snapshot image and prepare for another image processing, regardless of whether there are any specific features or Missing fe. In addition and instead, when detecting certain features, for example, at the open end of a gold-plated copper connector, no matter where # BGA is, it is preferable to generate a snapshot image and prepare it for other image processing, regardless of Place, or if -104-

536919 A7 B7 五、發明説明（ 102 發現有關於特色之任何特定缺陷。536919 A7 B7 V. Description of the Invention (102 Any specific defects regarding the features were found.

裝為了識別特色，可產生一 cel圖，其在圖40中所示之一部份，可使用供另外處理。cel圖可識別輪廓，及使每一輪廓與在其任一側存在一種材料相關，例如在一侧之鐘金銅及在另一侧之基片，電源線1012之結果亦將如此。 cel較佳為予以直接操作，以檢測凹隙1020及凸起部 1022 〇使用本發明，也產生一形態圖，圖40中示其一部份，其中產生一供物體之形態骨架。在形態圖，產生特色之骨架1024，並且骨架1024較佳為與一種特定材料相關。因此，例如在骨架1024上表示電源線1024之一部位之特色，將會與鍍金銅相關。訂In order to identify features, a cel chart can be generated, a portion of which is shown in FIG. 40, which can be used for additional processing. The cel diagram can identify the contours and associate each contour with the presence of a material on either side, such as a bell on one side and a copper substrate on the other side, as will the result of the power cord 1012. cel is preferably directly operated to detect the recess 1020 and the convex portion 1022. Using the present invention, a morphology diagram is also generated, and a part thereof is shown in FIG. 40, in which a morphological skeleton for an object is generated. In the morphology diagram, a characteristic skeleton 1024 is generated, and the skeleton 1024 is preferably related to a specific material. Therefore, for example, the characteristic of a part of the power line 1024 on the skeleton 1024 will be related to gold-plated copper. Order

請回頭參照圖1，如以上所說明，影像分析系統100較佳為可操作分析輸入影像，並傳遞分析之結果供另外處理。較佳為，如以上參照圖1-40所更詳細說明，供影像分析系統100之各組份，可供另外處理利用之分析結果另包括： 1. 資訊特徵說明缺陷：表面缺陷；凹隙及凸起部缺陷；以及有缺陷彩色輪廓元件（cel)。 2. 特徵說明所預期物品之資訊：二進制cel，一骨架，骨架特色，一彩色骨架，及彩色cel。 3. —快照，一般為包含繞每一所考慮之部位，例如繞每 -105- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) 536919 A7 B7 五、發明説明（ 103 一預定型式之特色，諸如電源線 1012及球 1010 (圖 40 )，每一使用者界定區域，及每一缺陷之區域之彩色 RGB快照。不限制上述之概括性，緊接上面之結果之每一清單，一般為在單獨之資料通道予以輸出，供依希望另外處理。較佳為，單獨之資料通道包含二實際通道如下：一第一通道，較佳為包括下列報告及資料：一二進制cel之報告；一彩色cel之報告；一形態及骨架特色之報告；以及一缺陷諸如凹隙及凸起部之報告。一第二通道較佳為包括下列報告及資料：一彩色缺陷報告，指示來自COMRADD 190之彩色表面缺陷，及來自CABS單元130之灰表面缺陷之報告；以及一快照報告，包括所考慮部位之快照，例如所報告缺陷之每一部位，每一所考慮之形態特色及每一所考慮之使用者界定之部位。請予察知，為求清楚起見，在單獨實施例之方面所說明之本發明之各不同特色，也可成組合提供在單一實施例。反之，為求簡便起見，在單一實施例之方面所說明之本發明之各不同特色，也可單獨或成任何適當子組合予以提供。 -106- 本紙張尺度適用中國國家標準(CNS) A4規格(210X297公釐)Please refer back to FIG. 1. As explained above, the image analysis system 100 is preferably operable to analyze the input image and pass the analysis result for additional processing. Preferably, as described above in more detail with reference to FIGS. 1-40, the analysis results for the components of the image analysis system 100 that can be used for additional processing include: 1. Information feature description defects: surface defects; recesses and Bump defects; and defective color contour elements (cel). 2. Feature Description Information of the expected item: binary cel, a skeleton, skeleton features, a colored skeleton, and colored cel. 3. —Snapshots, usually including around each considered part, such as around -105- This paper size applies Chinese National Standard (CNS) A4 specifications (210 X 297 mm) 536919 A7 B7 V. Description of the invention (103 The characteristics of a predetermined pattern, such as the power cord 1012 and ball 1010 (Figure 40), a color RGB snapshot of each user-defined area, and each defective area. Without limiting the above-mentioned generality, each of the above results immediately follows A list is generally output in a separate data channel for additional processing as desired. Preferably, a separate data channel contains two actual channels as follows: a first channel, preferably including the following reports and data: a binary cel A report of a color cel; a report of the shape and skeleton characteristics; and a report of defects such as dents and protrusions. A second channel preferably includes the following reports and information: a color defect report indicating the instructions from Reports of color surface defects in COMRADD 190 and gray surface defects from CABS unit 130; and a snapshot report, including a snapshot of the site under consideration, such as the report Each part of the defect, each considered morphological feature, and each considered user-defined part. Please be aware that for the sake of clarity, the various features of the invention are described in terms of separate embodiments It can also be provided in a single embodiment in combination. Conversely, for the sake of simplicity, the various features of the invention described in the aspect of a single embodiment can also be provided separately or in any appropriate sub-combination. -106- This paper size applies to China National Standard (CNS) A4 (210X297 mm)

裝 Φ 536919 A7 B7 五、發明説明（） 104 精於此項技藝者將會察知，本發明不受在上文曾特別圖示及說明者所限制。而是本發明之範圍係僅由後附申請專利範圍所界定： -107 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐)Installation Φ 536919 A7 B7 V. Description of the invention () 104 Those skilled in this art will know that the present invention is not limited by those who have been specifically illustrated and described above. Instead, the scope of the present invention is only defined by the scope of patents attached to the application: -107 This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm)

Claims

536919 Ϊ〇9^26ΐ專利申請案 BS 中又申印專利範圍替換本(92年2月）C8、申請專利範圍 Αδ B8 年月 {-4 修正 1· 一種表示有圖案物品之彩色圖案物品之形福之多色影：。’孩彩色影像包含有 2·如申請專利編1項之彩色影像，其中…. 色影像之部份之彩色指示在有 :、线多之材料。衣物口口之各不同部位圖案 3·如申請專利範圍第1或2嚷之彩色影像，並中有物品包含電路。電路包括含 4.如申請專利範圍冑3項之彩色影像，其中括導體之元件。電路包含印 5·如申請專利範圍第4項之彩色影像，其中刷電路板。 6·如申請專利㈣第4項之彩色影像，其中電路包含球狀格柵陣列基片。 7·如申請專利範則i或2項之彩色影像，纟中有圖案物品包含引線框架。 8.如申請專利範圍第丨或2項之彩色影像，進一步包括含括金屬鍍層之金屬塗層。 9·如申請專利範圍第6項之彩色影像，其中元件包含在球狀格栅陣列基片上之球體。 10. 如申請專利範圍第1或2項之彩色影像，其中有圖案物品包含蝕刻金屬基片。 11. 如申請專利範圍第1或2項之彩色影像，其中有圖案本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐) 536919 A B c D 六、申請專利範圍物品包含壓花金屬基片。 12. —種表示有圖案物品之彩色影像，該彩色影像包含一圖，其指示在物品之不同部位間之邊界，並在有共同邊界之每一不同部位識別材料。 13. 如申請專利範圍第1 2項之彩色影像，其中圖之部份之彩色指示在不同部位之材料。 14. 如申請專利範圍第1 3項之彩色影像，其中每一彩色包含一自預定彩色可能性之彩色。 15. 如申請專利範圍第12，13或14項之彩色影像，其中有圖案物品包含電路。 16. 如申請專利範圍第15項之彩色影像，其中電路包括含括導體之元件。 17. 如申請專利範圍第16項之彩色影像，其中電路包含印刷電路板。 18. 如申請專利範圍第16項之彩色影像，其中電路包含球狀格柵陣列基片。 19. 如申請專利範圍第12，13或14項之彩色影像，其中有圖案物品包含引線框架。 20. 如申請專利範圍第12，13或14項之彩色影像，進一步包括含括金屬鍍層之金屬塗層。 21. 如申請專利範圍第18項之彩色影像，其中元件包含在球狀格栅陣列基片上之球體。 -2- 本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐） 536919 A BCD 々、申請專利範圍 22. 如申請專利範圍第12，13或14項之彩色影像，其中有圖案物品包含蝕刻金屬基片。 23. 如申請專利範圍第12，13或14項之彩色影像，其中有圖案物品包含壓花金屬基片。本紙張尺度適用中國國家標準(CNS) A4規格(210 X 297公釐） 536919 第090117726號專利申請案中文圖式替換頁(92年2月） nr) s$s 卜麻芰改π 浹添蚌淞 S^5S 泠&> cps CELS 5S ^c^>許寒^^^:^^者^ 氐 s SES ^ss 碎536919 Ϊ〇9 ^ 26ΐ Patent application in BS and applied for replacement of the patent scope (February 1992) C8, the scope of patent application Aδ B8 Month {-4 Amendment 1 · The shape of a colored patterned article representing patterned articles Blessed multicolor shadow :. The color image of a child includes 2. • The color image of item 1 of the patent application, of which the color indication of the color image is in materials with multiple lines. Different parts of the clothing mouth pattern 3. As the color image of the patent application scope 1 or 21, there are items containing circuits. The circuit includes a color image containing 4. items such as the scope of patent application 胄 3, including the components of the conductor. The circuit contains a color image as in item 4 of the scope of patent application, in which the circuit board is brushed. 6. The color image of the fourth patent application, wherein the circuit includes a spherical grid array substrate. 7. If the color image of item i or item 2 of the patent application, the patterned item in the frame includes the lead frame. 8. The color image of item 丨 or 2 of the scope of patent application, further including a metal coating including a metal plating layer. 9. The color image according to item 6 of the patent application, wherein the element comprises a sphere on a spherical grid array substrate. 10. The color image of item 1 or 2 of the patent application, where the patterned article includes an etched metal substrate. 11. If the color image of item 1 or 2 of the patent application scope has a pattern, the paper size is applicable to Chinese National Standard (CNS) A4 specification (210 X 297 mm) 536919 AB c D Metal substrate. 12. A color image representing a patterned article, the color image including a picture indicating the boundaries between different parts of the article, and identifying the material at each different location with a common boundary. 13. If the color image of item 12 of the patent application scope, the color of the part of the figure indicates the material in different parts. 14. The color image of item 13 of the patent application, where each color contains a color from a predetermined color possibility. 15. For color images in the scope of patent application No. 12, 13 or 14, the patterned article contains the circuit. 16. For a color image in the scope of patent application No. 15, the circuit includes a component including a conductor. 17. The color image of claim 16 in which the circuit includes a printed circuit board. 18. A color image as claimed in claim 16, wherein the circuit includes a spherical grid array substrate. 19. If the color image of claim 12, 13 or 14 is applied for, the patterned article includes a lead frame. 20. The color image of claim 12, 13, or 14 further includes a metal coating including a metal plating. 21. The color image of claim 18, wherein the element comprises a sphere on a spherical grid array substrate. -2- This paper size applies to China National Standard (CNS) A4 specification (210 X 297 mm) 536919 A BCD 々, patent application scope 22. If the color image of the patent application scope item 12, 13 or 14, there are patterns The article includes an etched metal substrate. 23. If the color image of claim 12, 13, or 14, the patterned article includes an embossed metal substrate. This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) 536919 Patent Application No. 090117726 Chinese Schematic Replacement Page (February 1992) nr) s $ s 卜麻芰改 π 浃 Tim Clam淞 S ^ 5S && > cps CELS 5S ^ c ^ > Xu Han ^^^: ^^ 者 ^ 氐 s SES ^ ss broken

536919 9? 第090117726號專利申請案中文圖式替換頁(92年2月）圖39A 984536919 9? Patent Application No. 090117726 Chinese Schematic Replacement Page (February 1992) Figure 39A 984

圖39BFigure 39B

536919 09 第090117726號專利申請案中文圖式替換頁(92年2月） 1024 100 1014536919 09 Patent Application No. 090117726 Chinese Schematic Replacement Page (February 1992) 1024 100 1014

一 q一 5 1002 10? 1P 1000 a 4〇One q one 5 1002 10? 1P 1000 a 4〇